ubuntu下利用Dlib实现目标跟踪（下）——多目标跟踪

接着上一篇ubuntu下利用Dlib实现目标跟踪（上），本文继续介绍Dlib的目标跟踪功能。

用dlib来实现多多目标跟踪也是很容易的，就是多声明几个dlib.correlation_tracker()类就好了，将上一篇ubuntu下利用Dlib实现目标跟踪（上）中的代码加以修改，就可以达到多目标跟踪的效果了。

先贴代码，看效果：

import osimport globimport cv2import dlib# Path to the video framesvideo_folder = os.path.join("..", "examples", "video_frames2")# Create the correlation tracker - the object needs to be initialized before it can be usedtracker1 = dlib.correlation_tracker()tracker2 = dlib.correlation_tracker()tracker3 = dlib.correlation_tracker()selection = Nonetrack_window = Nonedrag_start = Nonedef onmouse(event, x, y, flags, param):    global selection,track_window,drag_start    if event == cv2.EVENT_LBUTTONDOWN:        drag_start = (x, y)        track_window = None    if drag_start:        xmin = min(x, drag_start[0])        ymin = min(y, drag_start[1])        xmax = max(x, drag_start[0])        ymax = max(y, drag_start[1])        selection = (xmin, ymin, xmax, ymax)    if event == cv2.EVENT_LBUTTONUP:        drag_start = None        track_window = selection        selection = Nonedef main():    track_window1 = ()    track_window2 = ()    track_window3 = ()    cv2.namedWindow('image',1)    cv2.setMouseCallback('image',onmouse)    # We will track the frames as we load them off of disk    for k, f in enumerate(sorted(glob.glob(os.path.join(video_folder, "*.jpg")))):        print("Processing Frame {}".format(k))        img_raw = cv2.imread(f)        image = img_raw.copy()        # We need to initialize the tracker on the first frame        if k == 0:            # Start a track on the object you want. box the object using the mouse and press 'Enter' to start tracking              while True:                img_first = image.copy()                if track_window:                    cv2.rectangle(img_first,(track_window[0],track_window[1]),(track_window[2],track_window[3]),(0,0,255),1)                elif selection:                    cv2.rectangle(img_first,(selection[0],selection[1]),(selection[2],selection[3]),(0,0,255),1)                if track_window1:                    cv2.rectangle(img_first,(track_window1[0],track_window1[1]),(track_window1[2],track_window1[3]),(0,255,255),1)                if track_window2:                    cv2.rectangle(img_first,(track_window2[0],track_window2[1]),(track_window2[2],track_window2[3]),(0,255,100),1)                if track_window3:                    cv2.rectangle(img_first,(track_window3[0],track_window3[1]),(track_window3[2],track_window3[3]),(200,0,200),1)                cv2.imshow('image',img_first)                if cv2.waitKey(10) == 10:                    if not track_window1:                        track_window1 = track_window                    elif not track_window2:                        track_window2 = track_window                    elif not track_window3:                        track_window3 = track_window                    else:                        break            tracker1.start_track(image, dlib.rectangle(track_window1[0], track_window1[1], track_window1[2], track_window1[3]))            tracker2.start_track(image, dlib.rectangle(track_window2[0], track_window2[1], track_window2[2], track_window2[3]))            tracker3.start_track(image, dlib.rectangle(track_window3[0], track_window3[1], track_window3[2], track_window3[3]))        else:            # Else we just attempt to track from the previous frame            tracker1.update(image)            tracker2.update(image)            tracker3.update(image)        # Get previous box and draw on showing image        box1_predict = tracker1.get_position()        box2_predict = tracker2.get_position()        box3_predict = tracker3.get_position()        cv2.rectangle(image,(int(box1_predict.left()),int(box1_predict.top())),(int(box1_predict.right()),int(box1_predict.bottom())),(0,255,255),1)        cv2.rectangle(image,(int(box2_predict.left()),int(box2_predict.top())),(int(box2_predict.right()),int(box2_predict.bottom())),(0,255,100),1)        cv2.rectangle(image,(int(box3_predict.left()),int(box3_predict.top())),(int(box3_predict.right()),int(box3_predict.bottom())),(200,0,200),1)        cv2.imshow('image',image)        cv2.waitKey(10)    cv2.destroyAllWindows()if __name__ == '__main__':    main()

注：video_frames2文件夹中放的是我Windows下利用dlib19.2实现多目标追踪 中用的视频图片集。

• 用鼠标框要跟踪的物体，框完按回车键确认，同样的手法连框三个
  

  

• 再次按回车键开始跟踪
  

  

跟踪效果还是很不错的

但发现dlib这个目标跟踪的物体如果遭遇遮挡，它就不灵光了（这个技术难度是很大的）

本文介绍的多目标跟踪方法和Windows下利用dlib19.2实现多目标追踪 中介绍的GitHub链接https://github.com/eveningglow/multi-object-tracker在效果上是差不多的，但速度快了的可不是一星半点儿。Windows下利用dlib19.2实现多目标追踪 也是用了Dlib和opencv，只是在Windows下用C++（不应该更快么）实现的，为什么有这么大的速度差别呢。

有空了再仔细研读一下人家的代码，看里面到底用到了什么黑科技。反正在本视频文件的跟踪效果上没有发现什么明显的差异。

---