Processing robbery case night time CCTV footage?

Question

I’ve got a cctv footage of bike robbery , video is not clear. I’m working on enhancing the video to find details about the vehicles or face or any other information from it. https://drive.google.com/file/d/0B8ZjiDWHZ1j-QkVKZFV4ZkV1ODA/view?usp=sharing originalhistogram equalisedCLAHE

What I've done till now in python opencv

Basic histogram equalisation because its a night time footage.
Since there were bright hotspots in the film, I applied contrast limited adaptive hist equalization (CLAHE) and experimented with cliplimit 1 to 10 and grid sizes from 4 to 32.
I've been playing with gabor filter to reduce the headlamp flash of the bikes but not able to tune it to the right parameters.
Since there is some noise in the video , thought of denoising it with fastNlDenoising but with no results.
Since image is small I tried super resolution of the image using SeRanet SeRanet on github.But its very slow to run on even a single image and results were not good enough to verify the face in the video. (at least on the frame I ran SeRanet over).

Any help on how can i proceed to gather more information from the video (or is it dead end by some limitations)

import numpy as np
import cv2
import time
cap = cv2.VideoCapture('./bike_robbery.mp4')
clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(32,32))
count=0
sig = 1
th = 0
lm = 1.0
gm = 0.02
ps = 0;
ksize =31
gaborKernel=cv2.getGaborKernel((ksize,ksize), sig, th, lm, gm, ps);
sleeptime=0.001
while(cap.isOpened()):
    ret, frame = cap.read()
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    cl1 = clahe.apply(gray)
    equ = cv2.equalizeHist(gray)
    #denoise=cv2.fastNlMeansDenoising(cl1,None,3,90,2)
    ths=cv2.threshold(equ, 230, 255, cv2.THRESH_BINARY)[1]
    gab=cv2.filter2D(cl1,0 , cv2.CV_32F, gaborKernel);
    kernel = np.array([[-1,-1,-1], [-1,18,-1], [-1,-1,-1]])
    im = cv2.filter2D(equ, -1, kernel)
    res = np.hstack((gray,equ,cl1)) #stacking images side-by-side
    cv2.imwrite("./clahe_pics/frame%d.jpg" % count, cl1)
    cv2.imwrite("./hist_pics/frame%d.jpg" % count, equ)
    count = count + 1
    cv2.imshow('frame',res)
    time.sleep(sleeptime)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break
    if cv2.waitKey(1) & 0xFF == ord('s'):
        sleeptime+=0.05
    if cv2.waitKey(1) & 0xFF == ord('f'):
        sleeptime-=.05
cap.release()
cv2.destroyAllWindows()

Processing robbery case night time CCTV footage?

1 Answers