在python中使用边界框更新OpenCV跟踪器

Question

I am using an OpenCV tracker to perform face tracking in videos, together with a face detector every few frames. If a face is detected by the face detector, I would like to update the tracker with the "detected" bounding box. I see that there is an option to enter a Rect in the C++ implementation but for some reason not in the python implementation as written in the opencv documentation . This is also an option when using dlib's correlation_tracker .

Currently, I can only initialize the tracker with a bounding box, but not update it with one in Python. if my tracker has drifted off the initial face it was tracking, I can't "bring it back" even if I know where the face is now (using my face detector). Is there a way to do this in python (eg should I kill the current tracker and init another with detected bounding box)?

Answer 1

I was looking for the very same thing and I found myself the solution to the problem by re-creating tracker each time successful detection occur. Please check following code. If something is not clear, feel free to ask for details:

import cv2 as cv

cap = cv.VideoCapture(0)

face_front_cascade = cv.CascadeClassifier("haarcascade_frontalface_alt.xml")    
tracker = cv.TrackerKCF_create()
bbox = ()

while True:
    ret,frame = cap.read()

    #press S to capture the face
    if cv.waitKey(20) & 0xFF == ord("s"):
        frame_gray = cv.cvtColor(frame,cv.COLOR_BGR2GRAY)
        face = face_front_cascade.detectMultiScale(frame_gray, scaleFactor=1.5, minNeighbors=3)
        for (x, y, w, h) in face: 
            colour = (0,0,255)
            stroke = 20
            cv.rectangle(frame,(x,y),(x+w,y+h), colour, stroke)
            bbox = (x,y,w,h)
            tracker = cv.TrackerKCF_create() #overwrite old tracker

    #trace face and draw box around it
    if bbox:
        tracker.init(frame, bbox)
        ret, bbox = tracker.update(frame)
        if ret:
            p1 = (int(bbox[0]), int(bbox[1]))
            p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
            cv.rectangle(frame, p1, p2, (255, 0, 0), 2, 1)

    #show result
    cv.imshow("frame",frame)

    #press ESC to exit
    if cv.waitKey(20) & 0xFF ==27:
        break    
cap.release()
cv.destroyAllWindows()