def draw_state(self, vis,vgg,images,sess, count1,frame_count,datadynamic):
(x, y), (w, h) = self.pos, self.size
x1, y1, x2, y2 = int(x-0.5*w), int(y-0.5*h), int(x+0.5*w), int(y+0.5*h)
cv2.rectangle(vis, (x1, y1), (x2, y2), (0, 0, 255))
if self.good:
cv2.circle(vis, (int(x), int(y)), 2, (0, 0, 255), -1)
checker = 0
else:
cv2.line(vis, (x1, y1), (x2, y2), (0, 0, 255))
cv2.line(vis, (x2, y1), (x1, y2), (0, 0, 255))
checker = 100
x1a = x1
y1a = y1
cropped = vis[y1:y1+y2 , x1:x1+x2]
img = cropped
x1, y1, z = img.shape
if x1 != 0 and y1 != 0 :
if(x1>y1):
x1 = (x1*224)/y1
y1 = 224
img_res = cv2.resize(img, (y1,int(x1)), interpolation = cv2.INTER_CUBIC)
x1= x1/2;
x1 = int (x1)
img_res = img_res[(x1)-112:(x1)+112, 0:224]
else:
y1 = int((y1*224)/x1)
x1 = 224
img_res = cv2.resize(img, (y1,x1), interpolation = cv2.INTER_CUBIC)
y1 = y1/2;
y1 = int (y1)
Y1 = y1 - 122
Y2 = y1 + 122
img_res = img_res[0:224 , Y1:Y2]
vgg1 = vg.imresize(img_res, (224, 224))
image_stack = np.stack([vgg1])
probs = sess.run(vgg.probs, feed_dict={vgg.imgs: image_stack})
preds = np.argmax(probs, axis=1)
for index, p in enumerate(preds):
cv2.putText(vis,"{}".format(class_names[p]),(x1a,y1a-20),cv2.FONT_HERSHEY_SIMPLEX,0.5,(0,0,255),2)
text = str(self.count1)
cv2.putText(vis,"{}".format(text),(x1a,y1a),cv2.FONT_HERSHEY_SIMPLEX,0.5,(0,0,255),2)
print("Dynamic,Frame Number:{}, Track Number:{},X-Axis:{}, Y-Axis:{}, Dynamic Prediction:{}, Probablity:{}".format(frame_count,text,x1,y1,class_names[p],probs[index,p]))
datadynamic=np.vstack((datadynamic,np.array(("Dynamic",frame_count,text,x1,y1,class_names[p],probs[index,p]))))
return checker,datadynamic
def run(self, vgg, images, sess, frame, cnts, count1, frame_count):
checker = 0
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
while True:
if not self.paused:
ret, self.frame = self.cap.read()
if not ret:
break
frame_gray = cv2.cvtColor(self.frame, cv2.COLOR_BGR2GRAY)
for tracker in self.trackers:
tracker.update(frame_gray)
vis = self.frame.copy()
(kp, descs) = surf.detectAndCompute(gray, None)
length = len(kp)
frame_list = np.zeros((length, 1), dtype=float)
for i in range(0, length):
frame_list[i, :] = kp[i].response
index = random.randint(0, length - 1)
xs = int(kp[index].pt[0])
ys = int(kp[index].pt[1])
val = int((kp[index].size) / 2)
new_val = val
#Resize Attention Window
if val < 50:
if xs < 100 or ys < 100 or xs > (width - 100) or ys > (
height - 100):
new_val = val
else:
new_val = 100
attention_window = vis[ys - new_val:ys + new_val,
xs - new_val:xs + new_val]
if len(self.trackers) > 0:
cv2.imshow('tracker state', self.trackers[-1].state_vis)
x0, y0, x1, y1 = self.rect
cv2.rectangle(vis, (xs - new_val, ys - new_val),
(xs + new_val, ys + new_val), (255, 0, 0), 3)
vgg1 = vg.imresize(attention_window, (224, 224))
image_stack = np.stack([vgg1])
probs = sess.run(vgg.probs, feed_dict={vgg.imgs: image_stack})
preds = np.argmax(probs, axis=1)
for index, p in enumerate(preds):
cv2.putText(vis, "{}".format(class_names[p]),
(xs - new_val, ys - new_val),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 2)
#print("Still,Frame Number:{},Frame Number:{},X-Axis:{}, Y-Axis:{}, Static Prediction:{}, Probablity:{}".format(frame_count,frame_count,xs,ys,class_names[p],probs[index,p]))
self.datastatic = np.vstack(
(self.datastatic,
np.array(("Still", frame_count, frame_count, xs, ys,
class_names[p], probs[index, p]))))
static_label = class_names[p]
for tracker in self.trackers:
(x, y), (w, h) = tracker.pos, tracker.size
x11, y11, x12, y12 = int(x - 0.5 * w), int(
y - 0.5 * h), int(x + 0.5 * w), int(y + 0.5 * h)
c, self.datadynamic = tracker.draw_state(
vis, vgg, images, sess, count1, frame_count,
self.datadynamic, x_new, x_old, xs, static_label)
if c == 100:
checker = 100
cv2.imshow('frame', vis)
fvid.write(vis)
ch = cv2.waitKey(10)
break
if ch == ord(' '):
self.paused = not self.paused
if ch == ord('c'):
self.trackers = []
return checker
def __init__(self, video_src, paused=False):
def run(self, vgg, images, sess, frame, cnts, count1, frame_count):
checker = 0
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
while True:
if not self.paused:
ret, self.frame = self.cap.read()
if not ret:
break
frame_gray = cv2.cvtColor(self.frame, cv2.COLOR_BGR2GRAY)
for tracker in self.trackers:
tracker.update(frame_gray)
vis = self.frame.copy()
(kp, descs) = surf.detectAndCompute(gray, None)
length = len(kp)
frame_list = np.zeros((length, 1), dtype=float)
for i in range(0, length):
frame_list[i, :] = kp[i].response
index = random.randint(0, length - 1)
xs = int(kp[index].pt[0])
ys = int(kp[index].pt[1])
val = int((kp[index].size) / 2)
new_val = val
#Resize Attention Window
if val < 50:
if xs < 100 or ys < 100 or xs > (width - 100) or ys > (
height - 100):
new_val = val
else:
new_val = 100
attention_window = vis[ys - new_val:ys + new_val,
xs - new_val:xs + new_val]
if len(self.trackers) > 0:
cv2.imshow('tracker state', self.trackers[-1].state_vis)
x0, y0, x1, y1 = self.rect
cv2.rectangle(vis, (xs - new_val, ys - new_val),
(xs + new_val, ys + new_val), (255, 0, 0), 3)
vgg1 = vg.imresize(attention_window, (224, 224))
image_stack = np.stack([vgg1])
probs = sess.run(vgg.probs, feed_dict={vgg.imgs: image_stack})
preds = np.argmax(probs, axis=1)
for index, p in enumerate(preds):
cv2.putText(vis, "{}".format(class_names[p]),
(xs - new_val, ys - new_val),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 2)
#print("Still,Frame Number:{},Frame Number:{},X-Axis:{}, Y-Axis:{}, Static Prediction:{}, Probablity:{}".format(frame_count,frame_count,xs,ys,class_names[p],probs[index,p]))
self.datastatic = np.vstack(
(self.datastatic,
np.array(("Still", frame_count, frame_count, xs, ys,
class_names[p], probs[index, p]))))
static_label = class_names[p]
for tracker in self.trackers:
(x, y), (w, h) = tracker.pos, tracker.size
x11, y11, x12, y12 = int(x - 0.5 * w), int(
y - 0.5 * h), int(x + 0.5 * w), int(y + 0.5 * h)
c, self.datadynamic = tracker.draw_state(
vis, vgg, images, sess, count1, frame_count,
self.datadynamic, x_new, x_old, xs, static_label)
if c == 100:
checker = 100
cv2.imshow('frame', vis)
fvid.write(vis)
ch = cv2.waitKey(10)
break
if ch == ord(' '):
self.paused = not self.paused
if ch == ord('c'):
self.trackers = []
return checker
#Initialisation
global fvid, lengthvid, frame_count
self.datastatic = []
self.datadynamic = []
x_new = np.zeros((10, 1))
x_old = np.zeros((10, 1))
fvid = cv2.VideoWriter('video.avi', 1, 24.0, (1280, 720))
p = 0
k = 0
count = 0
cnts = []
cnt1 = []
self.datastatic.append(
("Type", "Frame Number", "Frame Number", "X-Axis", "Y-Axis",
"Object Label", "Probablity"))
self.datadynamic.append(
("Type", "Frame Number", "Track Number", "X-Axis", "Y-Axis",
"Object Label", "Probablity"))
frame_count = 0
count1 = 0
count = 1
checker = 0
self.trackers = []
self.paused = paused
self.cap = cv2.VideoCapture('Tablee.mp4')
lengthvid = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT))
fgbg = cv2.createBackgroundSubtractorMOG2()
#Initialisation for VGG
sess = tf.Session()
images = tf.placeholder(tf.float32, [None, 224, 224, 3])
vgg = vg.vgg16(images, 'vgg16_weights.npz', sess)
stat_prob = np.zeros((5, 2), dtype=object)
stat_prob[0, 1] = 0.1
stat_prob[1, 1] = 0.11
stat_prob[2, 1] = 0.12
stat_prob[3, 1] = 0.13
stat_prob[4, 1] = 0.14
#Initialistaion for Static SURF
surf = cv2.xfeatures2d.SURF_create(1000, 5, 5)
width = self.cap.get(3)
height = self.cap.get(4)
frame_overlap = np.zeros((30, 3), dtype=float)
stat_count = 0
fcount = 0
while (1):
fcount += 1
_, self.frame = self.cap.read()
gray = cv2.cvtColor(self.frame, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (21, 21), 0)
fgmask = fgbg.apply(self.frame)
thresh = cv2.erode(fgmask, None, iterations=2)
(_, cnts, _) = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
count1 = 0
k = 0
lstatic = np.array(self.datastatic)
kstatic = lstatic[1:, 6]
zstatic = heapq.nlargest(1, kstatic)
ldynamic = np.array(self.datadynamic)
kdynamic = ldynamic[1:, 6]
zdynamic = heapq.nlargest(1, kdynamic)
key = cv2.waitKey(1) & 0xFF
for c in cnts:
if cv2.contourArea(c) < 2000:
continue
k = k + 1
if cnts != [] or k == 0:
if count == 1:
for c in cnts:
(x, y, w, h) = cv2.boundingRect(c)
self.rect = (x, y, x + w, y + h)
if cv2.contourArea(c) < 2000:
continue
p = p + 1
self.onrect(self.rect, count, count1)
count1 = count1 + 1
checker = run(self, vgg, images, sess, self.frame,
cnts, count1, frame_count)
count = count + 1
if checker == 100 or k != p:
for dat in self.datadynamic:
f1.write('|'.join(list(map(str, dat))) + '\n')
index_static = heapq.nlargest(1, range(len(kstatic)),
kstatic.__getitem__)
index_dynamic = heapq.nlargest(1, range(len(kdynamic)),
kdynamic.__getitem__)
print("{} is moving towards {}".format(
self.datadynamic[index_dynamic, 5],
self.datastatic[index_static, 5]))
checker = 0
self.trackers = []
p = 0
for c in cnts:
(x, y, w, h) = cv2.boundingRect(c)
self.rect = (x, y, x + w, y + h)
if cv2.contourArea(c) < 2000:
continue
p = p + 1
self.onrect(self.rect, count, count1)
count1 = count1 + 1
checker = run(self, vgg, images, sess, self.frame,
cnts, count1, frame_count)
count = 2
else:
checker = run(self, vgg, images, sess, self.frame, cnts,
count1, frame_count)
count1 = count1 + 1
else:
count1 = count1 + 1
(kp, descs) = surf.detectAndCompute(gray, None)
length = len(kp)
frame_list = np.zeros((length, 1), dtype=float)
for i in range(0, length):
frame_list[i, :] = kp[i].response
index = random.randint(0, length - 1)
xs = int(kp[index].pt[0])
ys = int(kp[index].pt[1])
val = int((kp[index].size) / 2)
new_val = val
#Resize Attention Window
if val < 50:
if xs < 100 or ys < 100 or xs > (width - 100) or ys > (
height - 100):
new_val = val
else:
new_val = 100
attention_window = self.frame[ys - new_val:ys + new_val,
xs - new_val:xs + new_val]
if len(self.trackers) > 0:
cv2.imshow('tracker state', self.trackers[-1].state_vis)
x0, y0, x1, y1 = self.rect
#cv2.rectangle(vis, (x0, y0), (x1, y1), (0, 255, 0), 3)
cv2.rectangle(self.frame, (xs - new_val, ys - new_val),
(xs + new_val, ys + new_val), (255, 0, 0), 3)
vgg1 = vg.imresize(attention_window, (224, 224))
image_stack = np.stack([vgg1])
probs = sess.run(vgg.probs, feed_dict={vgg.imgs: image_stack})
preds = np.argmax(probs, axis=1)
for index, p in enumerate(preds):
cv2.putText(self.frame, "{}".format(class_names[p]),
(xs - new_val, ys - new_val),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 2)
#print("Still,Frame Number:{},Frame Number:{},X-Axis:{}, Y-Axis:{}, Static Prediction:{}, Probablity:{}".format(frame_count,frame_count,xs,ys,class_names[p],probs[index,p]))
self.datastatic = np.vstack(
(self.datastatic,
np.array(("Still", frame_count, frame_count, xs, ys,
class_names[p], probs[index, p]))))
frame_count += 1
if key == ord("q"):
break
break
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
cars = car_cascade.detectMultiScale(gray, 1.1, 1)
for (x,y,w,h) in cars:
i=1
if ((w*h)<5000):
continue
else:
cv2.rectangle(img,(x,y),(x+w,y+h),(0,0,255),2)
car= img[y:y+h,x:x+w]
i+=1
vgg1 = vg.imresize(car, (224, 224))
#vgg2 = vg.imresize(car, (224, 224))
image_stack = np.stack([vgg1])
probs = sess.run(vgg.probs, feed_dict={vgg.imgs: image_stack})
preds = np.argmax(probs, axis=1)
for index, p in enumerate(preds):
cv2.putText(img,"{}".format(class_names[p]),(x,y),cv2.FONT_HERSHEY_SIMPLEX,0.5,(0,0,255),2)
for index, p in enumerate(preds):
print("Prediction: %s; Probability: %f"%(class_names[p], probs[index, p]))
cv2.imshow('video', img)
if cv2.waitKey(33) == 27:
break
cv2.destroyAllWindows()
