def __init__(self,vidfile,labelfile,modelFile,labelListFile,exportPath, perClassFrames=40,frameRate=20,context_size=5, banerWidth = 80,scale = 1): # input properties self.vidreader = VideoReader(vidfile); self.labelreader = open(labelfile,'r'); self.N = self.vidreader.frames; self.width,self.height = self.vidreader.width,self.vidreader.height; self.context_size=context_size; self.perClassFrames = perClassFrames; self.labels = load_labels(labelListFile); self.n_outs = len(self.labels); self.flag_colors = []; for index in range(self.n_outs): self.flag_colors.extend([tuple(np.array(cm.jet(index/float(self.n_outs))[:3][::-1])*255)]) self.predictors,self.predictorStartIdx = load_predictors(modelFile); self.input_shape = [self.height,self.width,3] self.batch_size = self.predictors[0].batch_size #write properites self.banerWidth = banerWidth self.vidWriter = VideoWriter(exportPath,self.banerWidth+self.width,self.height,fps=frameRate); self.colors = np.random.randint(256, size=(len(self.labels), 3)) self.scale = scale; #status self.frameIdx = 0; self.tasks = deque(); self.isFinished = False; self.vidWriter.build();
def write_video(write_path,
vidreader,
smoothMasks,
window_centers,
window_sizes,
num_blocks,
num_prev_frames,
rsz_shape=[80, 60]):
create_folder_structure_if_not_exists(write_path)
shape = [vidreader.width, vidreader.height]
vidwriter = VideoWriter(write_path, vidreader.width, vidreader.height)
vidwriter.build()
vidreader.__reset__()
skip_frames = num_prev_frames + num_blocks / 2
vidreader.skip_frames(skip_frames)
frame_idx = 0
numFrames = len(smoothMasks)
while (frame_idx < numFrames):
frame = vidreader.read_next()
if frame is None:
break
frame = draw_rect(frame, window_centers[frame_idx], window_sizes,
shape)
name = "test_results/tracking/track{0}.png".format(frame_idx)
cv2.imwrite(name, frame)
vidwriter.write(np.uint8(frame))
frame_idx += 1
vidwriter.close()
def tracker(inPath,outPath="test_results/tracker.avi"):
#initialization for default value
hog = cv2.HOGDescriptor()
hog.setSVMDetector( cv2.HOGDescriptor_getDefaultPeopleDetector())
vidreader = VideoReader(inPath)
vidwriter = VideoWriter(outPath,vidreader.width,vidreader.height)
vidwriter.build();
frame_idx =0; N = vidreader.frames;
while vidreader.num_remaining_frames() > 0:
frame_idx += 1;
cur_frame = vidreader.read_next();
print 'Perform human tracking.... {0}%\r'.format((frame_idx*100/N)),
if cur_frame is None:
break;
gray_cur_frame = cv2.cvtColor(cur_frame, cv2.COLOR_BGR2GRAY)
found, w = hog.detectMultiScale(gray_cur_frame, winStride=(8,8), padding=(32,32), scale=1.05)
found_filtered = []
#discarding the bounding box on within other
for ri, r in enumerate(found):
for qi, q in enumerate(found):
if ri != qi and inside(r, q):
break
else:
found_filtered.append(r)
draw_detections(cur_frame, found_filtered, 1)
vidwriter.write(cur_frame);
vidreader.close();
vidwriter.close()
cv2.destroyAllWindows()
print "Implemented Human Tracking............. [Done]"
def write_video(vidreader, out_path, num_prev_frames, num_blocks, smoothMasks,
vaMasks):
create_folder_structure_if_not_exists(out_path)
start = time.time()
vidwriter = VideoWriter(out_path, vidreader.width * 2, vidreader.height)
vidwriter.build()
vidreader.__reset__()
N = vidreader.frames
vidreader.skip_frames(num_prev_frames + num_blocks / 2)
frame_idx = 0
numFrames = len(smoothMasks)
while (frame_idx < numFrames):
print 'Writing video ... {0}%\r'.format((frame_idx * 100 / N)),
frame = vidreader.read_next()
zero_frame = np.zeros(frame.shape, dtype=np.uint8)
zero_frame[:, :, 2] = vaMasks[frame_idx] * 255
out_frame1 = cv2.addWeighted(frame, 0.6, zero_frame, 0.4, 0)
out_frame2 = frame * smoothMasks[frame_idx][:, :, None]
name = "test_results/smoothening/{0}_smooth.png".format(
frame_idx + num_prev_frames + num_blocks / 2)
cv2.imwrite(name, out_frame2)
vidwriter.write(np.uint8(np.hstack((out_frame1, out_frame2))))
frame_idx += 1
vidwriter.close()
time_taken = time.time() - start
print "Writing video .... [DONE] in ", time_taken, " seconds"
def createVideo(datapath,group,outfileName,shape,duration=120,framerate=20,vidLen=32):
vidOutFileName = outfileName + 'test.avi'
labelOutFileName = outfileName + 'test.txt'
values = [];
create_folder_structure_if_not_exists(vidOutFileName);
with open(datapath+os.sep+group+".txt",'r') as fp:
for line in fp:
values.extend([line.split()]);
numClass = len(values);
numExamples = (duration*framerate)/(vidLen);
randomNumbers = np.random.random_integers(0,numClass-1,numExamples);
fileWriter = open(labelOutFileName,'w');
vidWriter = VideoWriter(vidOutFileName,shape[0],shape[1]);
vidWriter.build();
for idx,random_idx in enumerate(randomNumbers):
print 'Creating UCF 50 video ... {0}%\r'.format((idx*100/numExamples)),
classDetails = values[random_idx];
classPath = datapath + os.sep + classDetails[0] + os.sep
classExamples = [os.path.join(classPath, _file) for _file in os.listdir(classPath)
if os.path.isfile(os.path.join(classPath, _file))];
chosenExample = classExamples[np.random.randint(len(classExamples))];
#print chosenExample
vidreader = VideoReader(chosenExample,None);
num_frames = vidreader.frames;
cnt,frames = vidreader.read(np.random.randint(num_frames-vidLen),vidLen);
fileWriter.writelines("%d\n" % int(item) for item in [classDetails[2]]*cnt)
for frame in frames:
vidWriter.write(frame);
vidWriter.close();
fileWriter.close();
def test_bgsub_mog(inp):
vidreader = VideoReader(inp)
vidwriter = VideoWriter("test_results/bg_sub_mog.avi", vidreader.width,
vidreader.height)
bgsub = get_instance(BGMethods.MOG_SUBSTRACTION)
start = time.time()
process_video(bgsub, vidreader, vidwriter)
time_taken = time.time() - start
print "Tested Background Subtraction (Mixture of Gaussian)... [DONE] in " + str(
time_taken) + " seconds"
def test_bgsub_fd(inp):
vidreader = VideoReader(inp)
vidwriter = VideoWriter("test_results/bg_sub_fd.avi", vidreader.width,
vidreader.height)
bgsub = get_instance(BGMethods.FRAME_DIFFERENCING)
start = time.time()
process_video(bgsub, vidreader, vidwriter, num_blocks=2)
time_taken = time.time() - start
print "Tested Background Subtraction (Frame Differencing)... [DONE] in " + str(
time_taken) + " seconds"
def createVideo(datapath,
group,
outfileName,
shape,
duration=120,
framerate=20,
vidLen=32):
vidOutFileName = outfileName + 'test.avi'
labelOutFileName = outfileName + 'test.txt'
values = []
create_folder_structure_if_not_exists(vidOutFileName)
with open(datapath + os.sep + group + ".txt", 'r') as fp:
for line in fp:
values.extend([line.split()])
numClass = len(values)
numExamples = (duration * framerate) / (vidLen)
randomNumbers = np.random.random_integers(0, numClass - 1, numExamples)
fileWriter = open(labelOutFileName, 'w')
vidWriter = VideoWriter(vidOutFileName, shape[0], shape[1])
vidWriter.build()
for idx, random_idx in enumerate(randomNumbers):
print 'Creating UCF 50 video ... {0}%\r'.format(
(idx * 100 / numExamples)),
classDetails = values[random_idx]
classPath = datapath + os.sep + classDetails[0] + os.sep
classExamples = [
os.path.join(classPath, _file) for _file in os.listdir(classPath)
if os.path.isfile(os.path.join(classPath, _file))
]
chosenExample = classExamples[np.random.randint(len(classExamples))]
#print chosenExample
vidreader = VideoReader(chosenExample, None)
num_frames = vidreader.frames
cnt, frames = vidreader.read(np.random.randint(num_frames - vidLen),
vidLen)
fileWriter.writelines("%d\n" % int(item)
for item in [classDetails[2]] * cnt)
for frame in frames:
vidWriter.write(frame)
vidWriter.close()
fileWriter.close()
def write_video(write_path,vidreader,smoothMasks,window_centers,window_sizes,num_blocks,
num_prev_frames,rsz_shape=[80,60]):
create_folder_structure_if_not_exists(write_path);
shape = [vidreader.width, vidreader.height];
vidwriter = VideoWriter(write_path,vidreader.width,vidreader.height);
vidwriter.build(); vidreader.__reset__();
skip_frames = num_prev_frames + num_blocks/2; vidreader.skip_frames(skip_frames);
frame_idx = 0; numFrames= len(smoothMasks);
while(frame_idx < numFrames):
frame = vidreader.read_next();
if frame is None:
break;
frame = draw_rect(frame,window_centers[frame_idx],window_sizes,shape);
name = "test_results/tracking/track{0}.png".format(frame_idx);
cv2.imwrite(name,frame);
vidwriter.write(np.uint8(frame))
frame_idx += 1;
vidwriter.close();
def write_video(vidreader,out_path,num_prev_frames,num_blocks,smoothMasks,vaMasks):
create_folder_structure_if_not_exists(out_path);
start = time.time();
vidwriter = VideoWriter(out_path,vidreader.width*2,vidreader.height)
vidwriter.build(); vidreader.__reset__(); N = vidreader.frames;
vidreader.skip_frames(num_prev_frames + num_blocks/2); frame_idx = 0
numFrames= len(smoothMasks);
while(frame_idx < numFrames):
print 'Writing video ... {0}%\r'.format((frame_idx*100/N)),
frame = vidreader.read_next();
zero_frame = np.zeros(frame.shape,dtype=np.uint8);
zero_frame[:,:,2] = vaMasks[frame_idx]*255
out_frame1 = cv2.addWeighted(frame,0.6,zero_frame,0.4,0);
out_frame2 = frame * smoothMasks[frame_idx][:,:,None]
name = "test_results/smoothening/{0}_smooth.png".format(frame_idx+num_prev_frames + num_blocks/2);
cv2.imwrite(name,out_frame2)
vidwriter.write(np.uint8(np.hstack((out_frame1,out_frame2))))
frame_idx += 1;
vidwriter.close();
time_taken = time.time()-start;
print "Writing video .... [DONE] in ",time_taken," seconds"
class UCF50Processor(object):
def __init__(self,vidfile,labelfile,modelFile,labelListFile,exportPath,
perClassFrames=40,frameRate=20,context_size=5,
banerWidth = 80,scale = 1):
# input properties
self.vidreader = VideoReader(vidfile);
self.labelreader = open(labelfile,'r');
self.N = self.vidreader.frames;
self.width,self.height = self.vidreader.width,self.vidreader.height;
self.context_size=context_size;
self.perClassFrames = perClassFrames;
self.labels = load_labels(labelListFile);
self.n_outs = len(self.labels);
self.flag_colors = [];
for index in range(self.n_outs):
self.flag_colors.extend([tuple(np.array(cm.jet(index/float(self.n_outs))[:3][::-1])*255)])
self.predictors,self.predictorStartIdx = load_predictors(modelFile);
self.input_shape = [self.height,self.width,3]
self.batch_size = self.predictors[0].batch_size
#write properites
self.banerWidth = banerWidth
self.vidWriter = VideoWriter(exportPath,self.banerWidth+self.width,self.height,fps=frameRate);
self.colors = np.random.randint(256, size=(len(self.labels), 3))
self.scale = scale;
#status
self.frameIdx = 0;
self.tasks = deque();
self.isFinished = False;
self.vidWriter.build();
def __design_frame_banner__(self,frame,_score,_label,top=3):
if not self.scale == 1:
frame = cv2.resize(frame,None,fx=self.scale, fy=self.scale, interpolation = cv2.INTER_CUBIC)
if frame.ndim == 2:
frame = np.dstack((frame,frame,frame));
assert(frame.ndim==3),"given frame not in shape"
baner_frame = np.zeros((self.height,self.banerWidth,3));
_indices = np.argsort(_score)[::-1];
col = 5; row = 8; steps = ((self.height-30)/(top+1))-5;
small_fface = cv2.FONT_HERSHEY_DUPLEX;
__draw_str__(baner_frame,(col+3,row),">PREDICTION<",color=(255,255,255),fontsize=0.25,fontface=small_fface); row += steps
for pos,classLbl in enumerate(_indices[:top]):
_str = "{0}. {1}".format(pos+1,self.labels[classLbl]);
__draw_str__(baner_frame,(col,row),_str,color=self.colors[classLbl],fontsize=0.25); row += steps
__draw_str__(baner_frame,(col+3,row),">ACTUAL<",color=(255,255,255),fontsize=0.25,fontface=small_fface); row += steps
if not _label is None:
_str = "{0}".format(self.labels[_label]);
__draw_str__(baner_frame,(col,row),_str,color=self.colors[_label],fontsize=0.25);
rank = list(_indices).index(_label); _str = "Rank : {0}".format(rank+1); row += steps;
__draw_str__(baner_frame,(col+3,row),_str,color = (255,255,255),fontsize=0.25);
cv2.rectangle(baner_frame,(8,self.height-12),(self.banerWidth-8,self.height-3),(255,255,255),1);
cv2.rectangle(baner_frame,(10,self.height-10),(self.banerWidth-10,self.height-5),self.flag_colors[rank],-1);
return np.hstack((baner_frame,frame));
def __readNextFrame__(self):
if self.vidreader.has_next():
frames = []; label = -1;
for idx in range(self.context_size):
frame = self.vidreader.read_next();
label = int(self.labelreader.readline());
frames.extend([frame]);
return (frames,label);
else:
return (None,-1);
def process(self):
blockReader = BlockReader(self.batch_size,self.input_shape,self.__readNextFrame__);
#pool = ThreadPool(processes = 2);
#task = pool.apply_async(self.__videoWriter__);
p_frames_cnt = 0
while not blockReader.isFinished:
(frameCnt,frames,labels)=blockReader.readNextBlock();
#print "READ A BLOCK...", len(self.tasks)
if frameCnt > 0:
processedWindowCenter = self.__process_block__(frames[:frameCnt]);
vidTask = BlockTask(frames,processedWindowCenter,labels,frameCnt)
#self.tasks.append(vidTask);
vidTask.process(self.vidWriter,self.predictors,self.__design_frame_banner__,self.predictorStartIdx);
self.isFinished = True;
def __process_block__(self,frameBlock):
windowCenters = [];
numGroupsPerClass = self.perClassFrames/self.context_size;
for idx in range(len(frameBlock)/numGroupsPerClass):
classFrames = [];
for frameGroup in frameBlock[idx*numGroupsPerClass:(idx+1)*numGroupsPerClass]:
for frame in frameGroup:
classFrames.extend([frame]);
c_windowCenters = processClassFrames(classFrames,self.context_size)
windowCenters.extend(c_windowCenters);
#print windowCenters
return windowCenters;
rect = np.array(
[_min[1], _min[0], _max[1] - _min[1], _max[0] - _min[0]],
dtype=np.float32)
cv2.accumulateWeighted(rect, prev_rect, 0.3, None)
rect = tuple(prev_rect)
cv2.rectangle(salProbColor, (rect[0], rect[1]),
(rect[0] + rect[2], rect[1] + rect[3]), (0, 0, 255))
mask_frame = np.float32(np.dstack((zero, mask * 255, zero)))
out_frame = cv2.addWeighted(np.float32(_frame), 0.6, mask_frame, 0.4,
0.0)
#out_frame = np.uint8( bg_es.cur_frame* mask[:,:,None])
frame = np.hstack((salProbColor, out_frame))
frame = cv2.resize(frame, (0, 0), fx=0.25, fy=0.25)
vidwriter.write(np.uint8(frame))
vidreader.close()
vidwriter.close()
if __name__ == "__main__":
import sys
if sys.argv.__len__() <= 1:
print "input path not provided........."
else:
inp = sys.argv[1]
out = "test_results/final.avi"
vidreader = VideoReader(inp)
vidwriter = VideoWriter(out, 2 * vidreader.width / 4,
vidreader.height / 4)
process(vidreader, vidwriter)
