def loop_and_detect(cam, trt_yolo, conf_th, vis, mjpeg_server):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolo: the TRT YOLO object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
mjpeg_server
"""
fps = 0.0
tic = time.time()
while True:
img = cam.read()
if img is None:
break
boxes, confs, clss = trt_yolo.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
mjpeg_server.send_img(img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
def loop_and_detect(cam, trt_yolo, conf_th, vis):
full_scrn = False
fps = 0.0
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is None:
break
boxes, confs, clss = trt_yolo.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
print('Inference Time: %s' % (toc - tic))
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
cv2.imwrite("trt_yolo_result.jpg", img)
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def img_callback(self, ros_img):
"""Continuously capture images from camera and do object detection """
tic = time.time()
# converts from ros_img to cv_img for processing
try:
cv_img = self.bridge.imgmsg_to_cv2(ros_img,
desired_encoding="bgr8")
rospy.logdebug("ROS Image converted for processing")
except CvBridgeError as e:
rospy.loginfo("Failed to convert image %s", str(e))
if cv_img is not None:
boxes, confs, clss = self.trt_yolo.detect(cv_img, self.conf_th)
cv_img = self.vis.draw_bboxes(cv_img, boxes, confs, clss)
toc = time.time()
fps = 1.0 / (toc - tic)
self.publisher(boxes, confs, clss)
if self.show_img:
cv_img = show_fps(cv_img, fps)
cv2.imshow("YOLOv4 DETECTION RESULTS", cv_img)
cv2.waitKey(1)
# converts back to ros_img type for publishing
try:
overlay_img = self.bridge.cv2_to_imgmsg(cv_img,
encoding="passthrough")
rospy.logdebug("CV Image converted for publishing")
self.overlay_pub.publish(overlay_img)
except CvBridgeError as e:
rospy.loginfo("Failed to convert image %s", str(e))
def loop_and_detect(cap, trt_yolo, result_video, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cap: the camera instance (video source).
trt_yolo: the TRT YOLO object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
#full_scrn = False
fps = 0.0
tic = time.time()
frame_width = int(cap.get(3))
frame_height = int(cap.get(4))
out = cv2.VideoWriter(result_video, cv2.VideoWriter_fourcc(*'mp4v'), 30,
(frame_width, frame_height))
while True:
ret, frame = cap.read()
boxes, confs, clss = trt_yolo.detect(frame, conf_th)
frame = vis.draw_bboxes(frame, boxes, confs, clss)
frame = show_fps(frame, fps)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
out.write(frame)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
cap.release()
out.release()
def loop_and_detect(cam, mtcnn, minsize):
"""Continuously capture images from camera and do face detection."""
full_scrn = False
fps = 0.0
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is not None:
dets, landmarks = mtcnn.detect(img, minsize=minsize)
print('{} face(s) found'.format(len(dets)))
img = show_faces(img, dets, landmarks)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps*0.95 + curr_fps*0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def loop_and_detect_console(cam, trt_ssd, conf_th, loop, cls_dict):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_ssd: the TRT SSD object detector instance.
conf_th: confidence/score threshold for object detection.
loop: run continuously
"""
fps = 0.0
tic = time.time()
run_loop = True # ensure we process the image once
while run_loop:
img = cam.read()
if img is not None:
boxes, confs, clss = trt_ssd.detect(img, conf_th)
print(list(zip(list(map(cls_dict.get, clss)), confs, boxes)),
end=" ")
img = show_fps(img, fps)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
print('fps={}'.format(curr_fps))
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
run_loop = loop
def loop_and_detect(cam, trt_yolov3, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolov3: the TRT YOLOv3 object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
full_scrn = False
fps = 0.0
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is not None:
time_start = time.time()
boxes, confs, clss = trt_yolov3.detect(img, conf_th)
print("--------------tf1----------\n",
time.time() - time_start, "\n-------------------------\n")
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def loop_and_display(condition):
"""Continuously capture images from camera and do classification."""
global s_img, s_probs, s_labels
full_scrn = False
fps = 0.0
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
with condition:
condition.wait()
img, top_probs, top_labels = s_img, s_probs, s_labels
show_top_preds(img, top_probs, top_labels)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('H') or key == ord('h'): # Toggle help message
show_help = not show_help
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def loop_and_detect(cam, trt_yolo, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolo: the TRT YOLO object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
full_scrn = False
fps = 0.0
tic = time.time()
while True:
# if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
# break
img = cam.read()
if img is None:
break
boxes, confs, clss = trt_yolo.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
# cv2.imshow(WINDOW_NAME, img)
print(f"{datetime.datetime.now().isoformat()} fps={fps}", flush=True)
cv2.imwrite("output.jpg", img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
def loop_and_detect(camera, trt_yolo, args, confidence_thresh, visual):
"""Continuously capture images from camera and do object detection.
# Arguments
camera: the camera instance (video source).
trt_yolo: the TRT YOLO object detector instance.
confidence_thresh: confidence/score threshold for object detection.
visual: for visualization.
"""
fps = 0.0
cumulative_frame_time = 0.0
iterations = 0
# endless loop when user provides single image/webcam
while len(camera.imageNames) != 0:
if args.activate_display and (cv2.getWindowProperty(WINDOW_NAME, 0) <
0):
break
img = camera.read()
if img is None:
break
tic = time.time()
boxes, confidences, classes = trt_yolo.detect(img, confidence_thresh)
toc = time.time()
if args.activate_display or args.write_images:
img = visual.draw_bboxes(img, boxes, confidences, classes)
img = show_fps(img, fps)
if args.activate_display:
cv2.imshow(WINDOW_NAME, img)
frame_time = (toc - tic)
cumulative_frame_time += frame_time
curr_fps = 1.0 / frame_time
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
if args.activate_display:
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
if args.write_images:
path = os.path.join(args.image_output,
os.path.basename(camera.currentImage))
print("Image path: ", path)
cv2.imwrite(path, img)
print("FPS: {:3.2f} and {} Images left.".format(
fps, len(camera.imageNames)))
append_coco(boxes, confidences, classes, camera)
iterations += 1
# Write coco json file when done
coco_file = json.dumps(resultJson, cls=NpEncoder)
f = open(args.result_json, "w+")
f.write(coco_file)
f.close()
print(f"Average FPS: {(1 / (cumulative_frame_time / iterations))}")
def check(trt_yolov3, conf_th, vis):
fps = 0.0
tic = time.time()
counter = 0
capt = cv2.VideoCapture(ADDR_IN)
width = int(capt.get(cv2.CAP_PROP_FRAME_WIDTH ))
height = int(capt.get(cv2.CAP_PROP_FRAME_HEIGHT ))
fps = int(capt.get(cv2.CAP_PROP_FPS))
dimension = '{}x{}'.format(width, height)
command = ['ffmpeg',
'-y',
'-f', 'rawvideo',
'-vcodec','rawvideo',
'-pix_fmt', 'bgr24',
'-s', dimension,
'-i', '-',
'-c:v', 'libx264',
'-pix_fmt', 'yuv420p',
# '-preset', 'veryfast',
'-f', 'flv',
ADDR_OUT]
proc = sp.Popen(command, stdin=sp.PIPE, shell=False)
while True:
f, img = capt.read()
if img is not None:
# img = cv2.resize(img, (416, 416))
boxes, confs, clss = trt_yolov3.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps*0.95 + curr_fps*0.05)
tic = toc
counter += 1
proc.stdin.write(img.tostring())
capt.release()
capt.stop()
proc.stdin.close()
proc.stderr.close()
proc.wait()
def loop_and_detect(cam, trt_yolo, conf_th, vis):
full_scrn = False
tic = time.time()
#while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is None:
break
boxes, confs, clss = trt_yolo.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
inferenceTime = toc - tic
print('Inference Time: %s' % (inferenceTime))
def detect_demo(cam, trt_yolov3, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolov3: the TRT YOLOv3 object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
full_scrn = False
fps = 0.0
tic = time.time()
img = cam.read()
if img is not None:
boxes, confs, clss = trt_yolov3.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
img = show_fps(img, curr_fps)
cv2.imwrite("demo.png", img)
def loop_and_detect(cam, trt_yolo, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolo: the TRT YOLO object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
full_scrn = False
fps = 0.0
tic = time.time()
while True:
# if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
# break
img = cam.read()
if img is None:
break
boxes, confs, clss = trt_yolo.detect(img, conf_th)
# result=np.stack((clss,confs),axis=1)
# result=np.column_stack((result,boxes))
# print("The number of insulator we detected :%d"%len(result))
# print("The confidence of detection result: ",confs)
# print("The class of detection result: ",dic_class[clss[0]])
# print("The bounding box of detection result:\n",boxes)
#
# print(result)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imwrite("./result.jpg", img)
# cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
key = cv2.waitKey(1)
if fps > 0: # ESC key: quit program
break
def loop_and_detect(cam, trt_yolov3, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolov3: the TRT YOLOv3 object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
fourcc = cv2.VideoWriter_fourcc(*'XVID')
fps = cam.get(cv2.CAP_PROP_FPS)
size = (int(cam.get(cv2.CAP_PROP_FRAME_WIDTH)), int(cam.get(cv2.CAP_PROP_FRAME_HEIGHT)))
out = cv2.VideoWriter('camera_test.avi', fourcc, fps, size)
full_scrn = False
fps = 0.0
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is not None:
boxes, confs, clss = trt_yolov3.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps*0.95 + curr_fps*0.05)
tic = toc
#img = cv2.putText(img, "fps= %.2f" % (fps), (0, 40), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
out.write(img)
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def loop_and_display(condition, vis):
"""Take detection results from the child thread and display.
# Arguments
condition: the condition variable for synchronization with
the child thread.
vis: for visualization.
"""
global s_img, s_boxes, s_confs, s_clss
full_scrn = False
fps = 0.0
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
with condition:
# Wait for the next frame and detection result. When
# getting the signal from the child thread, save the
# references to the frame and detection result for
# display.
if condition.wait(timeout=MAIN_THREAD_TIMEOUT):
img, boxes, confs, clss = s_img, s_boxes, s_confs, s_clss
else:
raise SystemExit('ERROR: timeout waiting for img from child')
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def loop_and_detect(cam, trt_retinaface, priors, cfg):
"""Continuously capture images from camera and do face detection.
# Arguments
cam: the camera instance (video source).
trt_retinaface: the TRT_RetinaFace face detector instance.
priors: priors boxes with retinaface model
cfg: retinaface model parameter configure
"""
full_scrn = False
fps = 0.0
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is None:
break
facePositions, landmarks = trt_retinaface.detect(priors, cfg, img)
for (x1, y1, x2, y2), landmark in zip(facePositions, landmarks):
cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255), 2)
cv2.circle(img, (landmark[0], landmark[1]), 1, (0, 0, 255), 2)
cv2.circle(img, (landmark[2], landmark[3]), 1, (0, 255, 255), 2)
cv2.circle(img, (landmark[4], landmark[5]), 1, (255, 0, 255), 2)
cv2.circle(img, (landmark[6], landmark[7]), 1, (0, 255, 0), 2)
cv2.circle(img, (landmark[8], landmark[9]), 1, (255, 0, 0), 2)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps*0.95 + curr_fps*0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def loop_and_detect(cam, mtcnn, minsize):
"""Continuously capture images from camera and do face detection."""
full_scrn = False
fps = 0.0
tic = time.time()
start = tic
count = 0
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is not None:
dets, landmarks = mtcnn.detect(img, minsize=minsize)
print('{} face(s) found'.format(len(dets)))
count += 1
#img = show_faces(img, dets, landmarks)
img = show_fps(img, fps)
for bb, ll in zip(dets, landmarks):
x1, y1, x2, y2 = int(bb[0]), int(bb[1]), int(bb[2]), int(bb[3])
crop_img = img[y1:y2, x1:x2]
cv2.imshow(WINDOW_NAME, crop_img)
toc = time.time()
if count == 300:
print(f'Found 300 faces in {toc-start} seconds')
print(f'Average FPS: {300/(toc-start)}')
break
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def loop_and_detect(cam, trt_yolov3, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolov3: the TRT YOLOv3 object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
full_scrn = False
fps = 0.0
tic = time.time()
counter = 0
capt = cv2.VideoCapture(ADDR)
while True:
# if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
# break
img = capt.read()
img = cv2.resize(img, (416, 416))
print(img.shape)
if img is not None:
boxes, confs, clss = trt_yolov3.detect(img, conf_th)
print('detect')
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
# cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps*0.95 + curr_fps*0.05)
tic = toc
counter += 1
cv2.imwrite('/home/shared_folder/' + str(counter) + '.png', img)
def loop_and_detect(cam, trt_ssd, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_ssd: the TRT SSD object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
full_scrn = False
fps = 0.0
tic = time.time()
second = 1
count = 0
ave_p = 0
ave_bi = 0
p_count = 0
bi_count = 0
id_num, date_str, time_str, m = get_time()
prev_time = int(m / 10)
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is None:
break
boxes, confs, clss = trt_ssd.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
count = count + 1
if count >= int(fps):
count = 0
second = second + 1
p, bi = counting(clss)
ave_p = ave_p + p
ave_bi = ave_bi + bi
print(p, bi, ave_p, ave_bi, p_count, bi_count)
if second % 11 == 0:
second = 1
p_count = p_count + ave_p / 10
bi_count = bi_count + ave_bi / 3
ave_p = ave_bi = 0
id_num, date_str, time_str, m = get_time()
print(id_num, date_str, time_str, m, prev_time)
if prev_time != int(m / 10):
insert('seongbuk',
[id_num, date_str, time_str,
int(p_count),
int(bi_count)])
prev_time = int(m / 10)
p_count = bi_count = ave_p = ave_bi = second = 0
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
#print('result:',result)
height = frame.shape[0]
width = frame.shape[1]
if len(clss) == 0:
continue
else:
det = result[:, 0:5]
print('det:', det)
#det[:, 0] = det[:, 0] * width
#det[:, 1] = det[:, 1] * height
#det[:, 2] = det[:, 2] * width
#det[:, 3] = det[:, 3] * height
#print('det1',det)
trackers = mot_tracker.update(det)
frame = show_fps(frame, fps)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 +
curr_fps * 0.05)
tic = toc
for d in trackers:
xmin = int(d[0])
ymin = int(d[1])
xmax = int(d[2])
ymax = int(d[3])
label = int(d[4])
cv2.rectangle(frame, (xmin, ymin), (xmax, ymax),
(0, 0, 255), 2)
def loop_and_detect(cam, trt_yolo, conf_th, vis):
full_scrn = False
tic = time.time()
#while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
img = cam.read()
if img is None:
boxes, confs, clss = trt_yolo.detect(img, conf_th)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
inferenceTime = toc - tic
print ('Inference Time: %s' % (inferenceTime))
def main():
args = parse_args()
if args.category_num <= 0:
raise SystemExit('ERROR: bad category_num (%d)!' % args.category_num)
if not os.path.isfile('yolo/%s.trt' % args.model):
raise SystemExit('ERROR: file (yolo/%s.trt) not found!' % args.model)
cam = Camera(args)
if not cam.isOpened():
raise SystemExit('ERROR: failed to open camera!')
cls_dict = get_cls_dict(args.category_num)
yolo_dim = args.model.split('-')[-1]
if 'x' in yolo_dim:
dim_split = yolo_dim.split('x')
if len(dim_split) != 2:
raise SystemExit('ERROR: bad yolo_dim (%s)!' % yolo_dim)
w, h = int(dim_split[0]), int(dim_split[1])
else:
h = w = int(yolo_dim)
if h % 32 != 0 or w % 32 != 0:
raise SystemExit('ERROR: bad yolo_dim (%s)!' % yolo_dim)
start_load_time = time.time()
trt_yolo = TrtYOLO(args.model, (h, w), args.category_num)
stop_load_time = time.time()
load_time = stop_load_time - start_load_time
print ('Load Time: %s' % (load_time))
open_window(
WINDOW_NAME, 'Camera TensorRT YOLO Demo',
cam.img_width, cam.img_height)
vis = BBoxVisualization(cls_dict)
loop_and_detect(cam, trt_yolo, conf_th=0.3, vis=vis)
cam.release()
cv2.destroyAllWindows()
if __name__ == '__main__':
main()
cap = cv2.VideoCapture(0) #("rtsp://*****:*****@192.168.10.67:554/1/1")
cls_dict = get_cls_dict('coco')
yolo_dim = int(args.model.split('-')[-1]) # 416 or 608
trt_yolov3 = TrtYOLOv3(args.model, (yolo_dim, yolo_dim))
vis = BBoxVisualization(cls_dict)
full_scrn = False
fps = 0.0
tic = time.time()
conf_th = 0.3
while True:
_, frame = cap.read()
time_start = time.time()
boxes, confs, clss = trt_yolov3.detect(frame, conf_th)
print("--------------tf1----------\n",
time.time() - time_start, "\n-------------------------\n")
img = vis.draw_bboxes(frame, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
def loop_and_detect(cam, trt_yolo, conf_th, vis, \
cls_dict, client, topic):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolo: the TRT YOLO object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
full_scrn = False
fps = 0.0
tic = time.time()
frame_id = 0
frame_id_holder = []
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is None:
break
boxes, confs, clss = trt_yolo.detect(img, conf_th)
print("boxes:", boxes)
print("confs:", confs)
print("clss:", clss)
#print(len(confs))
#for cl in clss:
#print("clss:",cl)
#print("clss:",clss)
#publish_bboxes(boxes, confs, clss, cls_dict, client, topic)
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
## wasuremono logic
# no detected person and object or one or more persons is only detected
# do not perform abandoned object detection
if len(clss) == 0 or ((len(clss) > 0) and (sum(clss) == 0)):
#print("length_confs:",len(confs))
print("no detected person and object or person is only detected")
pass
# only objects are detected
# perform abandoned object detection
elif 0 not in clss:
print("only objects are detected")
if (24 in clss) or (25 in clss) or (26 in clss) or (
28 in clss) or (67 in clss):
## Here needs to be added the function of the abandonment ##
add_centroid_classes_metrics2 = compute_pixel_distance(
boxes, clss, cls_dict)
#print("add_centroid_classes_metrics:", add_centroid_classes_metrics2)
for row in add_centroid_classes_metrics2:
if (row[6]=='backpack') or (row[6]=='umbrella') or (row[6]=='handbag') or \
(row[6]=='suitcase') or (row[6]=='cell phone'):
x_min2 = int(float(row[0]))
y_min2 = int(float(row[1]))
x_max2 = int(float(row[2]))
y_max2 = int(float(row[3]))
print("add_centroid_classes_metrics2:", row)
color = (0, 0, 255)
cv2.rectangle(img, (x_min2, y_min2), (x_max2, y_max2),
color, 2)
cv2.putText(img, 'abandoned', (x_min2 + 1, y_min2 - 2),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0),
2)
publish_bboxes_New(row, 'abandoned', client, topic)
# one or more persons and objects are detected
# perform abandoned object detection
# except the center of object is inside a detected person's bounding box
else:
print("one or more persons and objects are detected")
## Here needs to be added the function of the abandonment ##
#compute_pixel_per_metric_ratio(boxes, clss, cls_dict)
#extract_pixel_per_metric_ratio = compute_pixel_per_metric_ratio(boxes, clss, cls_dict)
#print("extract_pixel_per_metric_ratio:",extract_pixel_per_metric_ratio[-2][-2:])
add_centroid_classes_metrics = compute_pixel_distance(
boxes, clss, cls_dict)
print()
for row in add_centroid_classes_metrics:
if 'person' == row[6]:
coord_person_min_x = int(float(row[0]))
coord_person_min_y = int(float(row[1]))
coord_person_max_x = int(float(row[2]))
coord_person_max_y = int(float(row[3]))
for objectX in add_centroid_classes_metrics:
if 'person' != objectX[6]:
centroid_object_x = int(float(objectX[4]))
centroid_object_y = int(float(objectX[5]))
if (coord_person_min_x < centroid_object_x < coord_person_max_x) and \
(coord_person_min_y < centroid_object_y < coord_person_max_y):
pass
else:
dis_min = 100000
for temp_min in objectX[
7:len(add_centroid_classes_metrics) +
6]:
if float(temp_min) < float(dis_min):
dis_min = float(temp_min)
print()
print("dis_min:", dis_min)
if dis_min > 300:
print("")
print("objectX:", objectX)
print("wasuremono no kanouseiari")
if (frame_id >
(fps *
AOD_config.INITIAL_INSPECTION_DURATION
)):
frame_id_holder.append(frame_id)
if (objectX[6]=='backpack') or (objectX[6]=='umbrella') or (objectX[6]=='handbag') or \
(objectX[6]=='suitcase') or (objectX[6]=='cell phone'):
# (B, G, R)
color = (0, 255, 255)
x_min2 = int(float(objectX[0]))
y_min2 = int(float(objectX[1]))
x_max2 = int(float(objectX[2]))
y_max2 = int(float(objectX[3]))
#cv2.rectangle(img, (100,100), (200,200), color, 2)
cv2.rectangle(
img, (x_min2, y_min2),
(x_max2, y_max2), color, 2)
cv2.putText(
img, 'warning',
(x_min2 + 1, y_min2 - 2),
cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(0, 0, 0), 2)
print("frame_id_holder:",
frame_id_holder)
if frame_id >= (fps * (
(min(frame_id_holder) / fps) +
AOD_config.
ABANDONMENT_DURATION)):
color = (0, 0, 255)
cv2.rectangle(
img, (x_min2, y_min2),
(x_max2, y_max2), color, 2)
cv2.putText(
img, 'abandoned',
(x_min2 + 1, y_min2 - 2),
cv2.FONT_HERSHEY_SIMPLEX,
0.5, (0, 0, 0), 2)
publish_bboxes_New(
objectX, 'abandoned',
client, topic)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
print("fps:", fps)
tic = toc
frame_id += 1
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
def loop_and_detect(cam, trt_yolo, msg_queue, conf_th, vis):
fps = 0.0
resList = []
frame_cnt = 0
tic = time.time()
pv_detection = []
full_scrn = False
capture = cv2.VideoCapture(VIDEOPATH)
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
ret, frame = capture.read()
if frame is None:
break
frame_h, frame_w = frame.shape[:-1]
if frame_cnt > START_FRAME:
if frame_cnt == START_FRAME + 1:
matching_area = frame
boxes, confs, clss = trt_yolo.detect(matching_area, conf_th)
boxes, confs, clss = person_filter(
boxes, confs,
clss) # boxes --np.array() [[p, p, p, p, index], ...]
if frame_cnt > START_FRAME + 1:
if len(boxes) == 0:
# print("boxes after_filter is empty!")
boxes, confs, clss = pv_detection[:]
else:
for i in range(len(boxes)):
boxes[i][0:2] = np.add(
boxes[i][0:2], np.array(matching_area_top_left))
boxes[i][2:4] = np.add(
boxes[i][2:4], np.array(matching_area_top_left))
# 绘图
if frame_cnt == START_FRAME + 1:
img = vis.draw_bboxes(frame, boxes, confs, clss, 0, 0)
else:
img = vis.draw_bboxes(frame, boxes, confs, clss,
matching_area_top_left,
matching_area_bottom_right)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
# nominate desirable object
if frame_cnt == START_FRAME + 1:
print("Select the target you want.")
nmn = int(cv2.waitKey(0)) - 176
box = boxes[int(np.argwhere(boxes[:, -1] == int(nmn))), :
-1] # box --np.array() [p, p, p, p]
else:
box = find_best_box(boxes, frame, template)
#串口数据 定点数转换
ltx, lty, rbx, rby = box[:4] # ltx --left top point x
shift_x = int(
round(((ltx + rbx) - frame_w) / (2 * frame_w), 4) * 10000)
shift_y = int(
round(((lty + rby) - frame_h) / (2 * frame_h), 4) * 10000)
area_ratio = int(
round((rbx - ltx) * (rby - lty) / (frame_w * frame_h), 4) *
10000)
ges = -1
msg = str(shift_x) + ', ' + str(shift_y) + ', ' + str(
area_ratio) + ', ' + str(ges) + ';'
print("msg in python:", msg)
if msg_queue.empty():
msg_queue.put(msg.encode()) # 将字符转换为字节发送
roi = frame[box[1]:box[3], box[0]:box[2]]
if frame_cnt == START_FRAME + 1:
template = roi.copy()
roi_h, roi_w = roi.shape[:-1]
if roi_h < int(0.5 * frame_h):
roi_h = int(0.5 * frame_h)
if roi_w < int(0.2 * frame_w):
roi_w = int(0.2 * frame_w)
matching_area_top_left = [0, 0]
matching_area_bottom_right = [0, 0]
matching_area_top_left[0] = box[0] - int(0.5 * roi_w)
matching_area_top_left[1] = box[1] - int(0.25 * roi_h)
# apply kalman filter
if frame_cnt > START_FRAME + 20:
matching_area_top_left_measurement = matching_area_top_left
matching_area_top_left = kalman_prediction(
[[matching_area_top_left[0]], [matching_area_top_left[1]]])
# print(np.subtract(np.array(matching_area_top_left), np.array(matching_area_top_left_measurement)))
else:
dump = kalman_prediction([[matching_area_top_left[0]],
[matching_area_top_left[1]]])
matching_area_bottom_right[0] = box[2] + int(0.5 * roi_w)
matching_area_bottom_right[1] = box[3] + int(0.25 * roi_h)
# 越界处理
for i in range(len(matching_area_top_left)):
if matching_area_top_left[i] < 0:
matching_area_top_left[i] = 0
if matching_area_bottom_right[0] > frame.shape[1]:
matching_area_bottom_right[0] = frame.shape[1]
if matching_area_bottom_right[1] > frame.shape[0]:
matching_area_bottom_right[1] = frame.shape[0]
# 切片 [高, 宽]
matching_area = frame[
matching_area_top_left[1]:matching_area_bottom_right[1],
matching_area_top_left[0]:matching_area_bottom_right[0]]
# cv2.imshow("matching_area", matching_area)
# 人脸检测部分 速度超慢
# gray = cv2.cvtColor(matching_area, code=cv2.COLOR_BGR2GRAY)
# face_zone = face_detector.detectMultiScale(gray, scaleFactor = 1.1, minNeighbors = 5)
# resList.append(len(face_zone))
# if frame_cnt > START_FRAME + 5:
# preRes=ges
# ges = judge(resList, preRes)
# if 1 == ges:
# cv2.putText(matching_area, "front", (10,10), cv2.FONT_HERSHEY_COMPLEX, 0.5, (132, 255, 255), 1)
# else:
# cv2.putText(matching_area, "back", (10,10), cv2.FONT_HERSHEY_COMPLEX, 0.5, (255, 255, 152), 1)
# else:
# preRes = 0
# ges = 0
#
# f_num = 0
# for x, y, w, h in face_zone:
# f_num = f_num + 1
# cv2.rectangle(matching_area, pt1 = (x, y), pt2 = (x+w, y+h), color = [255,255,255], thickness=1)
# cv2.putText(matching_area, str(f_num), (x,y), cv2.FONT_HERSHEY_COMPLEX, 1, (0, 255, 0), 1)
# cv2.putText(matching_area, "{}people".format(f_num), (10,50), cv2.FONT_HERSHEY_COMPLEX, 1, (142, 125, 52), 1)
# cv2.imshow('result', matching_area) # 这里由于显示在frame上有点麻烦 所以直接show出来了
# calculate an exponentially decaying average of fps number
toc = time.time()
curr_fps = 1.0 / (toc - tic)
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
pv_detection = [boxes, confs, clss]
# cv2.waitKey(0)
key = cv2.waitKey(1)
if key == 27 or key == ord(' '):
break
elif key == ord('F') or key == ord('f'):
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
frame_cnt += 1
def loop_and_detect(cam, trt_yolo, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolo: the TRT YOLO object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
time_tmp = 0
count_t = [0, 0]
flag = [0, 0]
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out_video = cv2.VideoWriter('/home/ee201511400/test5.mp4', fourcc, 30.0,
(cam.img_width, cam.img_height))
full_scrn = False
fps = 0.0
tic = time.time()
people_num = 0
id_position = []
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is None:
break
resize = cv2.resize(img, (640, 480))
img_temp = img
boxes, confs, clss, no_mask_count, mask_count = trt_yolo.detect(
img, conf_th)
std_box_class, _clss = get_sorted_box(boxes, clss)
img_e = cv2.resize(img,
dsize=(WIDTH, HEIGHT),
interpolation=cv2.INTER_AREA)
people_num, id_position, time_for_disinfection = execute(
img_e, img, people_num, id_position, std_box_class, cam)
# handling all the return values
if no_mask_count == 0:
if (len(time_for_disinfection) == 2):
for index, el in enumerate(time_for_disinfection):
if (el == 1):
count_t[index] += 1
elif (el == 0):
count_t[index] = 0
else:
count_t[index] = count_t[index]
for index, el in enumerate(count_t):
if (el >= 20):
flag[index] = 1
print("time_count", count_t)
for index, el in enumerate(flag):
if (len(id_position) is not 0 and id_position[index] is not 0
and people_num == 1):
if (el == 1):
if (index == 1):
cv2.putText(img, "You are allowed to enter",
(400, 50), font, 1.0, (32, 32, 32), 1,
line)
else:
cv2.putText(img, "You are allowed to enter",
(200, 50), font, 1.0, (32, 32, 32), 1,
line)
else:
if (index == 1):
cv2.putText(img, "You are not allowed to enter",
(400, 50), font, 1.0, (32, 32, 32), 1,
line)
else:
cv2.putText(img, "You are not allowed to enter",
(200, 50), font, 1.0, (32, 32, 32), 1,
line)
elif (len(std_box_class) == 2):
if (el == 1):
cv2.putText(img, "You are allowed to enter",
(std_box_class[index][0], 50), font, 1.0,
(32, 32, 32), 1, line)
else:
cv2.putText(img, "You are not allowed to enter",
(std_box_class[index][0], 50), font, 1.0,
(32, 32, 32), 1, line)
else:
cv2.putText(img, "Put on your mask!", (50, 300), font, 2.0,
(0, 0, 255), 2, line)
if people_num == 0:
count_t = [0, 0]
flag = [0, 0]
cv2.putText(img, "Now you can come in", (50, 300), font, 2.0,
(255, 0, 0), 2, line)
img = vis.draw_bboxes(img, boxes, confs, _clss)
img = show_fps(img, fps)
cv2.putText(img, 'mask = ' + str(mask_count), (11, 300), font, 3.0,
(32, 32, 32), 4, line)
cv2.putText(img, 'mask = ' + str(mask_count), (10, 300), font, 3.0,
(240, 240, 240), 1, line)
cv2.putText(img, 'peaple = ' + str(people_num), (11, 400), font, 3.0,
(32, 32, 32), 4, line)
cv2.putText(img, 'peaple = ' + str(people_num), (10, 400), font, 3.0,
(240, 240, 240), 1, line)
people_num = 0
cv2.imshow(WINDOW_NAME, img)
out_video.write(img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
print('FPS = %d\n' % (fps))
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
f.close()
out_video.release()
def loop_and_detect(cam, trt_yolov3, conf_th, vis):
"""Continuously capture images from camera and do object detection.
# Arguments
cam: the camera instance (video source).
trt_yolov3: the TRT YOLOv3 object detector instance.
conf_th: confidence/score threshold for object detection.
vis: for visualization.
"""
full_scrn = False
fps = 0.0
detectandspeech = np.zeros((6, 2), dtype=float)
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is not None:
boxes, confs, clss = trt_yolov3.detect(img, conf_th)
###########################
## DETECTED CASE - SSMIN ##
###########################
if len(clss) > 0:
print("Detected : ", clss, ", ", confs)
detectandspeech[clss, 0] = tic
if detectandspeech[clss, 0] - detectandspeech[clss, 1] > 60:
detectandspeech[clss, 1] = tic
print("Played : ", clss)
MusicPlayCheck(clss)
if clss == 1:
os.system('mpg321 voice/jason.mp3 & > /dev/null')
elif clss == 2:
os.system('mpg321 voice/jessica.mp3 & > /dev/null')
elif clss == 3:
os.system('mpg321 voice/erica.mp3 & > /dev/null')
elif clss == 4:
os.system('mpg321 voice/woo.mp3 & > /dev/null')
elif clss == 5:
os.system('mpg321 voice/woong.mp3 & > /dev/null')
###########################
draw = 1
if draw is not 0:
img = vis.draw_bboxes(img, boxes, confs, clss)
img = show_fps(img, fps)
cv2.imshow(WINDOW_NAME, img)
toc = time.time()
curr_fps = 1.0 / (toc - tic)
# calculate an exponentially decaying average of fps number
fps = curr_fps if fps == 0.0 else (fps * 0.95 + curr_fps * 0.05)
tic = toc
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
elif key == ord('F') or key == ord('f'): # Toggle fullscreen
full_scrn = not full_scrn
set_display(WINDOW_NAME, full_scrn)
time.sleep(0.05)
def main():
# initialize camera class
cam = Camera()
if not cam.is_opened():
raise SystemExit('EROR: failed to open camera!')
cls_dict = alprClassNames()
# Yolo dimensions (416x416)
yolo_dim = 416
h = w = int(yolo_dim)
# Initialized model and tools
cwd = os.getcwd()
model_yolo = str(cwd) + '/weights/yolov4-tiny-416.trt'
model_crnn = str(cwd) + '/weights/crnn.pth'
trt_yolo = TrtYOLO(model_yolo, (h,w), category_num=1) # category number is number of classes
crnn = alpr.AutoLPR(decoder='bestPath', normalise=True)
crnn.load(crnn_path=model_crnn)
open_window(WINDOW_NAME, TITLE, cam.img_width, cam.img_height)
vis = BBoxVisualization(cls_dict)
# Loop and detect
full_scrn = False
fps = 0.0
tic = time.time()
while True:
if cv2.getWindowProperty(WINDOW_NAME, 0) < 0:
break
img = cam.read()
if img is None:
break
# Detect car plate
boxes confs, clss = trt_yolo.detect(img, conf_th=0.5)
# Crop and preprocess car plate
cropped = vis.crop_plate(img, boxes, confs, clss)
# Recognize car plate
lp_plate = ''
fileLocate = str(cwd) + '/detections/detection1.jpg'
if os.path.exists(fileLocate):
lp_plate = lpr.predict(fileLocate)
# Draw boxes and fps
img = vis.draw_bboxes(img, boxes, confs, clss, lp=lp_plate)
img = show_fps(img, fps)
# Show image
cv2.imshow(WINDOW_NAME, img)
# Calculate fps
toc = time.time()
curr_fps = 1.0 / (toc - tic)
fps = curr_fps if fps == 0.0 else (fps*0.95 + curr_fps*0.05)
tic = toc
# Exit key
key = cv2.waitKey(1)
if key == 27: # ESC key: quit program
break
# Release capture and destroy all windows
cam.release()
cv2.destroyAllWindows()