def index(request):
decodeddata = request.body.decode('utf-8')
dictdata = ast.literal_eval(decodeddata)
username = dictdata["username"]
imagename = dictdata["imagename"]
imageurl = dictdata["imageurl"]
start_time = time.time()
# Final TextBox++ Code : (Works on just image)
input_size = input_shape[:2]
print(input_size)
# getting the image
url = imageurl
response = requests.get(url)
img = Image.open(BytesIO(response.content))
img = np.array(img)
img_h = img.shape[0]
img_w = img.shape[1]
img1 = np.copy(img)
img2 = np.zeros_like(img)
# model to predict
x = np.array([preprocess(img, input_size)])
elapsed_time = time.time() - start_time
print("Performace measure : " + str(elapsed_time))
#Model start
start_time = time.time()
with sl_graph.as_default():
with sl_session.as_default():
y = sl_model.predict(x)
elapsed_time = time.time() - start_time
print("Performace measure : " + str(elapsed_time))
#Model end
start_time = time.time()
result = prior_util.decode(y[0], confidence_threshold)
if len(result) > 0:
bboxs = result[:, 0:4]
quads = result[:, 4:12]
rboxes = result[:, 12:17]
boxes = np.asarray([rbox3_to_polygon(r) for r in rboxes])
xy = boxes
xy = xy * [img_w, img_h]
xy = np.round(xy)
xy = xy.astype(np.int32)
cv2.polylines(img1, tuple(xy), True, (0, 0, 255))
rboxes = np.array(
[polygon_to_rbox(b) for b in np.reshape(boxes, (-1, 4, 2))])
bh = rboxes[:, 3]
rboxes[:, 2] += bh * 0.1
rboxes[:, 3] += bh * 0.2
boxes = np.array([rbox_to_polygon(f) for f in rboxes])
boxes = np.flip(boxes, axis=1) # TODO: fix order of points, why?
boxes = np.reshape(boxes, (-1, 8))
boxes_mask_a = np.array([b[2] > b[3] for b in rboxes
]) # width > height, in square world
boxes_mask_b = np.array([
not (np.any(b < 0) or np.any(b > 512)) for b in boxes
]) # box inside image
boxes_mask = np.logical_and(boxes_mask_a, boxes_mask_b)
boxes = boxes[boxes_mask]
rboxes = rboxes[boxes_mask]
xy = xy[boxes_mask]
if len(boxes) == 0:
boxes = np.empty((0, 8))
# draw
saveimageindjango = 'assets/mloutput_' + username + "_" + imagename
cv2.imwrite(saveimageindjango, img1)
elapsed_time = time.time() - start_time
print("Performace measure : " + str(elapsed_time))
print("Sending to back end...")
files = {'file': open(saveimageindjango, 'rb')}
headers = {
'username': username,
}
response = requests.request("POST",
'http://localhost:4000/upload',
files=files,
headers=headers)
print(response)
print("Backend Process Complete")
context = {"data": "data"}
return render(request, 'index.html', context)
def detect_motion(frameCount):
# lock variables
global vs, outputFrame, lock
# loop over frames from the video stream and edit anything here...
while True:
# read the next frame from the video stream, resize it,
# convert the frame to grayscale, and blur it
ret, frame = cap.read()
print("READING FRAME")
if frame is not None:
# model to predict
img = np.array(frame)
img_h = img.shape[0]
img_w = img.shape[1]
img1 = np.copy(img)
img2 = np.zeros_like(img)
# model to predict
x = np.array([preprocess(img, input_size)])
#Model start
start_time = time.time()
with sl_graph.as_default():
with sl_session.as_default():
y = sl_model.predict(x)
#Model end
result = prior_util.decode(y[0], confidence_threshold)
if len(result) > 0:
bboxs = result[:, 0:4]
quads = result[:, 4:12]
rboxes = result[:, 12:17]
boxes = np.asarray([rbox3_to_polygon(r) for r in rboxes])
xy = boxes
xy = xy * [img_w, img_h]
xy = np.round(xy)
xy = xy.astype(np.int32)
cv2.polylines(img1, tuple(xy), True, (0, 0, 255))
rboxes = np.array([
polygon_to_rbox(b) for b in np.reshape(boxes, (-1, 4, 2))
])
bh = rboxes[:, 3]
rboxes[:, 2] += bh * 0.1
rboxes[:, 3] += bh * 0.2
boxes = np.array([rbox_to_polygon(f) for f in rboxes])
boxes = np.flip(boxes,
axis=1) # TODO: fix order of points, why?
boxes = np.reshape(boxes, (-1, 8))
boxes_mask_a = np.array([b[2] > b[3] for b in rboxes
]) # width > height, in square world
boxes_mask_b = np.array([
not (np.any(b < 0) or np.any(b > 512)) for b in boxes
]) # box inside image
boxes_mask = np.logical_and(boxes_mask_a, boxes_mask_b)
boxes = boxes[boxes_mask]
rboxes = rboxes[boxes_mask]
xy = xy[boxes_mask]
if len(boxes) == 0:
boxes = np.empty((0, 8))
top = 10
bottom = 10
left = 10
right = 10
total_transcript = ""
# draw fps
frame = img1
# acquire the lock, set the output frame, and release the
# lock
with lock:
outputFrame = frame.copy()
# model to predict
x = np.array([preprocess(img, input_size)])
with sl_graph.as_default():
with sl_session.as_default():
y = sl_model.predict(x)
result = prior_util.decode(y[0], confidence_threshold)
if len(result) > 0:
bboxs = result[:, 0:4]
quads = result[:, 4:12]
rboxes = result[:, 12:17]
boxes = np.asarray([rbox3_to_polygon(r) for r in rboxes])
xy = boxes
xy = xy * [vid_w, vid_h]
xy = np.round(xy)
xy = xy.astype(np.int32)
cv2.polylines(img1, tuple(xy), True, (0, 0, 255))
rboxes = np.array([
polygon_to_rbox(b) for b in np.reshape(boxes, (-1, 4, 2))
])
bh = rboxes[:, 3]
rboxes[:, 2] += bh * 0.1
rboxes[:, 3] += bh * 0.2
boxes = np.array([rbox_to_polygon(f) for f in rboxes])
def detect_motion(frameCount):
# lock variables
global vs, outputFrame, lock
# loop over frames from the video stream and edit anything here...
while True:
# read the next frame from the video stream, resize it,
# convert the frame to grayscale, and blur it
previousCoordinates = ""
peopleindex = 0
peoplemapping = {}
strPeopleMapping = ""
ret, frame = cap.read()
print("READING FRAME")
if frame is not None:
# yolo
resultyolo = tfnet.return_predict(frame)
# model to predict
img = np.array(frame)
img_h = img.shape[0]
img_w = img.shape[1]
img1 = np.copy(img)
img2 = np.zeros_like(img)
# model to predict
x = np.array([preprocess(img, input_size)])
#Model start
start_time = time.time()
with sl_graph.as_default():
with sl_session.as_default():
y = sl_model.predict(x)
#Model end
result = prior_util.decode(y[0], confidence_threshold)
if len(result) > 0:
bboxs = result[:,0:4]
quads = result[:,4:12]
rboxes = result[:,12:17]
boxes = np.asarray([rbox3_to_polygon(r) for r in rboxes])
xy = boxes
xy = xy * [img_w, img_h]
xy = np.round(xy)
xy = xy.astype(np.int32)
cv2.polylines(img1, tuple(xy), True, (0,0,255))
rboxes = np.array([polygon_to_rbox(b) for b in np.reshape(boxes, (-1,4,2))])
bh = rboxes[:,3]
rboxes[:,2] += bh * 0.1
rboxes[:,3] += bh * 0.2
boxes = np.array([rbox_to_polygon(f) for f in rboxes])
boxes = np.flip(boxes, axis=1) # TODO: fix order of points, why?
boxes = np.reshape(boxes, (-1, 8))
boxes_mask_a = np.array([b[2] > b[3] for b in rboxes]) # width > height, in square world
boxes_mask_b = np.array([not (np.any(b < 0) or np.any(b > 512)) for b in boxes]) # box inside image
boxes_mask = np.logical_and(boxes_mask_a, boxes_mask_b)
boxes = boxes[boxes_mask]
rboxes = rboxes[boxes_mask]
xy = xy[boxes_mask]
if len(boxes) == 0:
boxes = np.empty((0,8))
top = 10
bottom = 10
left = 10
right = 10
total_transcript = ""
# To get the cropped out boxes and run pytesseract over it
for i in xy:
crop_img = img1[i[0][1]-5:i[2][1]+5,i[0][0]-5:i[2][0]+5]
color = [255,255,255]
crop_img = cv2.copyMakeBorder(crop_img, top, bottom, left, right, cv2.BORDER_CONSTANT,value=color)
transcript = pytesseract.image_to_string(crop_img, lang='eng').upper()
total_transcript += transcript + "\n"
print(transcript)
print(total_transcript)
# draw fps
frame = img1
# Start yolo process here
currentCoordinates = ""
# textbox++
img = frame
img_h = img.shape[0]
img_w = img.shape[1]
img1 = np.copy(img)
coordinates = previousCoordinates.split("\n")
coordinates.pop()
# YOLO-9000 : Drawing Boxes
peopleCount = 0
for res in resultyolo:
if res["label"] == "whole":
continue
elif res["label"] != "person":
color = int(255 * res["confidence"])
top = (res["topleft"]["x"], res["topleft"]["y"])
bottom = (res["bottomright"]["x"], res["bottomright"]["y"])
# for each person
cv2.rectangle(frame, top, bottom, (255,0,0) , 2)
cv2.putText(frame, res["label"], top, cv2.FONT_HERSHEY_DUPLEX, 1.0, (0,0,255))
elif res["label"] == "person":
peopleCount = peopleCount + 1
color = int(255 * res["confidence"])
top = (res["topleft"]["x"], res["topleft"]["y"])
bottom = (res["bottomright"]["x"], res["bottomright"]["y"])
topstr = "("+str(res["topleft"]["x"]) + \
","+str(res["topleft"]["y"])+")"
bottomstr = "("+str(res["bottomright"]["x"]) + \
","+str(res["bottomright"]["y"])+")"
coordinatesStr = {}
coordinatesStr['x1'] = top[0]
coordinatesStr['x2'] = bottom[0]
coordinatesStr['y1'] = top[1]
coordinatesStr['y2'] = bottom[1]
currentValue = topstr+" "+bottomstr
# IOU PART - BEGIN
currentCoordinates = currentCoordinates+topstr+" "+bottomstr+"\n"
# Calculate IoU here with top and bottom, compare each drawn image with top and bottom, select the max IoU
if previousCoordinates != "":
bb2 = {}
bb2['x1'] = top[0]
bb2['x2'] = bottom[0]
bb2['y1'] = top[1]
bb2['y2'] = bottom[1]
currentIou = 0
iouIndex = 0
for currentIndex, boxes in enumerate(coordinates):
boxesarr = boxes.split(" ")
top = ast.literal_eval(boxesarr[0])
bottom = ast.literal_eval(boxesarr[1])
bb1 = {}
bb1['x1'] = top[0]
bb1['x2'] = bottom[0]
bb1['y1'] = top[1]
bb1['y2'] = bottom[1]
result = get_iou(bb1, bb2)
temp = currentIou
currentIou = max(result, currentIou)
if temp != currentIou:
iouIndex = currentIndex
if currentIou != 0:
peoplemapping[currentValue] = peoplemapping[coordinates[iouIndex]]
# check for index:
try:
if peoplemapping[currentValue]:
pass
except:
peopleindex = peopleindex + 1
peoplemapping[currentValue] = peopleindex
else:
try:
if peoplemapping[currentValue]:
pass
except:
peopleindex = peopleindex + 1
peoplemapping[currentValue] = peopleindex
# IOU PART - END
strPeopleMapping = strPeopleMapping+currentValue+":"+str(peoplemapping[currentValue])+"|"
cv2.rectangle(img1,(coordinatesStr['x1'],coordinatesStr['y1']),(coordinatesStr['x2'],coordinatesStr['y2']), (255,0,0) , 2)
cv2.putText(img1,"index : "+str(peoplemapping[currentValue]),(coordinatesStr['x1'],coordinatesStr['y1']),cv2.FONT_HERSHEY_DUPLEX,1.0,(0,0,255))
frame = img1
previousCoordinates = currentCoordinates
strPeopleMapping = strPeopleMapping+"\n"
# acquire the lock, set the output frame, and release the
# lock
with lock:
outputFrame = frame.copy()