OpenCV 中的 KLT 跟踪器与 Python 无法正常工作

Question

I am using KLT (Kanade-Lucas-Tomasi Tracking) Tracking algorithm to track the motion of traffic in India. I am tracking flow of one side of traffic properly, but other side of traffic, that is moving in frame is not detected at all.

Algorithm consist of cv2.goodFeaturesToTrack and cv2.calcOpticalFlowPyrLK to achieve the result.

In the image you can see that Red and Silver car have no tracking feature on them. Yellow Auto on left is also not tracked. Any reason for this? Corners are still there.

Feature Parameters for cv2.goodFeaturesToTrack :

feature_params = dict( maxCorners = 500,   # How many pts. to locate
                       qualityLevel = 0.1,  # b/w 0 & 1, min. quality below which everyone is rejected
                       minDistance = 7,   # Min eucledian distance b/w corners detected
                       blockSize = 3 ) # Size of an average block for computing a derivative covariation matrix over each pixel neighborhood

Feature Parameters for cv2.calcOpticalFlowPyrLK :

lk_params = dict( winSize  = (15,15),  # size of the search window at each pyramid level
                  maxLevel = 2,   #  0, pyramids are not used (single level), if set to 1, two levels are used, and so on
                  criteria = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 0.03))

Video I have to use is 60 min. long and KLT stops tracking after 5 min .. Any suggestions or help would be great. Thanks.

Answer 1

Basically you are doing everything right you just need to reinitialize the good points for tracking like this

p0 = cv2.goodFeaturesToTrack(old_gray, mask = None, **feature_params)

after say every 5th frame or whatever you like Hope it helps ! following is my code:

import cv2
import numpy as np

cap = cv2.VideoCapture('side.avi')
# params for ShiTomasi corner detection
feature_params = dict( maxCorners = 100,
                       qualityLevel = 0.3,
                       minDistance = 7,
                       blockSize = 7 )
# Parameters for lucas kanade optical flow
lk_params = dict( winSize  = (15,15),
                  maxLevel = 2,
                  criteria = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 0.03))
# Create some random colors
color = np.random.randint(0,255,(100,3))
# Take first frame and find corners in it
ret, old_frame = cap.read()
for i in range(60):
    ret, old_frame = cap.read()
old_gray = cv2.cvtColor(old_frame, cv2.COLOR_BGR2GRAY)
p0 = cv2.goodFeaturesToTrack(old_gray, mask = None, **feature_params)
print(p0)
# Create a mask image for drawing purposes
mask = np.zeros_like(old_frame)
while(1):
    ret,frame = cap.read()
    frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    frame_no = cap.get(cv2.CAP_PROP_POS_FRAMES)
    if int(frame_no)%5 == 0:
        p0 = cv2.goodFeaturesToTrack(old_gray, mask = None, **feature_params)
    # calculate optical flow
    p1, st, err = cv2.calcOpticalFlowPyrLK(old_gray, frame_gray, p0, None, **lk_params)
    # Select good points
    good_new = p1[st==1]
    good_old = p0[st==1]
    # draw the tracks
    for i,(new,old) in enumerate(zip(good_new,good_old)):
        a,b = new.ravel()
        c,d = old.ravel()
        mask = cv2.line(mask, (a,b),(c,d), color[i].tolist(), 2)
        frame = cv2.circle(frame,(a,b),5,color[i].tolist(),-1)
    img = cv2.add(frame,mask)
    cv2.imshow('frame',img)
    k = cv2.waitKey(2000) & 0xff
    if k == 27:
        break
    # Now update the previous frame and previous points
    old_gray = frame_gray.copy()
    p0 = good_new.reshape(-1,1,2)
cv2.destroyAllWindows()
cap.release()

Answer 2

import numpy as np
import cv2

video_path = ''
output_file = ""     
cap = cv2.VideoCapture(video_path)

fourcc = cv2.VideoWriter_fourcc(*'DIVX')

# params for ShiTomasi corner detection
feature_params = dict( maxCorners = 500,   # How many pts. to locate
                       qualityLevel = 0.1,  # b/w 0 & 1, min. quality below which everyone is rejected
                       minDistance = 7,   # Min eucledian distance b/w corners detected
                       blockSize = 3 ) # Size of an average block for computing a derivative covariation matrix over each pixel neighborhood

# Parameters for lucas kanade optical flow
lk_params = dict( winSize  = (15,15),  # size of the search window at each pyramid level
                  maxLevel = 2,   #  0, pyramids are not used (single level), if set to 1, two levels are used, and so on
                  criteria = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 0.03))

''' Criteria : Termination criteria for iterative search algorithm.
    after maxcount { Criteria_Count } : no. of max iterations.
    or after { Criteria Epsilon } : search window moves by less than this epsilon '''


# Take first frame and find corners in it
ret, old_frame = cap.read()
old_gray = cv2.cvtColor(old_frame, cv2.COLOR_BGR2GRAY)
p0 = cv2.goodFeaturesToTrack(old_gray, mask=None, **feature_params)  #use goodFeaturesToTrack to find the location of the good corner.

# Create a mask image for drawing purposes filed with zeros
mask = np.zeros_like(old_frame)

y = 0
is_begin = True # To save the output video
count = 1  # for the frame count
n = 50  # Frames refresh rate for feature generation

while True:
    ret,frame = cap.read()
    if frame is None:
        break
    processed = frame

    #Saving the Video
    if is_begin:
        h, w, _ = processed.shape
        out = cv2.VideoWriter(output_file, fourcc, 30, (w, h), True)
        is_begin = False

    # Convert to Grey Frame
    frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

    if count%n == 0:  # Refresh the tracking features after every 50 frames
        cv2.imwrite('img/r{0:05d}.jpg'.format(y), img)
        y += 1
        ret, old_frame = cap.read()
        old_gray = cv2.cvtColor(old_frame, cv2.COLOR_BGR2GRAY)
        p0 = cv2.goodFeaturesToTrack(old_gray, mask=None, **feature_params)
        mask = np.zeros_like(old_frame)

    # calculate optical flow
    p1, st, err = cv2.calcOpticalFlowPyrLK(old_gray, frame_gray, p0, None, **lk_params)

    # Select good points
    good_new = p1[st==1]
    good_old = p0[st==1]

    # draw the tracks
    for i,(new,old) in enumerate(zip(good_new,good_old)):
        a,b = new.ravel() #tmp new value
        c,d = old.ravel() #tmp old value
        #draws a line connecting the old point with the new point
        mask = cv2.line(mask, (a,b),(c,d), (0,255,0), 1)
        #draws the new point
        frame = cv2.circle(frame,(a,b),2,(0,0,255), -1)
    img = cv2.add(frame,mask)

    out.write(img)
    cv2.imshow('frame',img)
    k = cv2.waitKey(30) & 0xff

    #Show the Output
    if k == 27:
        cv2.imshow('', img)
        break

    # Now update the previous frame and previous points
    old_gray = frame_gray.copy()
    p0 = good_new.reshape(-1,1,2)

    count += 1

# release and destroy all windows
cv2.destroyAllWindows()
cap.release()

I added the refresh rate for the GoodFeaturetoTrack and it's working, but we will not get the full trajectory. Working on it now.