Object detection in videos

Python
1

I have been trying to find an efficient way to detect common objects in a video.

Could someone please suggest if this could be done with the built-in modules in opencv or some other, more trusted library?

This is my code as of now. However, the cvlib library does not seem to be a good option.

import cv2
import cvlib
from gtts import gTTS
from playsound import playsound
labels=[]
def speech(text):
    print(text)
    language='en'
    output=gTTS(text=text,lang=language,slow=False)
    output.save('/Users/advikajain/Documents/OpenCv work/sounds/output.mp3')
    playsound('/Users/advikajain/Documents/OpenCv work/sounds/output.mp3')

video=cv2.VideoCapture('/Users/advikajain/Documents/OpenCv work/videos/test2.mp4')
while (video.isOpened()):
    ret,frame=video.read()
    bbox,label,conf=cvlib.detect_common_objects(frame)
    output=cvlib.object_detection.draw_bbox(frame,bbox,label,conf)
    cv2.imshow("advika",output)
    for item in label:
        if item in labels:
            pass
        else:
            labels.append(item)
    if cv2.waitKey(1) & 0xff == ord(' '):
        break

i=0
new_sentence=[]
for label in labels:
    if i==0:
        new_sentence.append(f"I found a {label},and,")
    else:
        new_sentence.append(f"a {label},")
    
    i+=1

speak=speech(str(new_sentence))
5

apart from not being maintained from opencv, – why so ?
if there are problems, describe in more detail, please.

what exactly ? are those covered by the 80 COCO classes ?

6

Yes. They are covered by the 80 COCO classes.
The CVlib library has only 1 maintainer and it has not been updated in quite a while.

7
1 Like
10

just to leave a note here:
if you think, the cvlib code above is ‘loosing overlapping detections’, it’s probably because NMS filtering should be done per class (as it is done in darknet & opencv) , not on the whole output set