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]"
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;