OpenCV-yolo-stream

Real time object detection in (youtube) video stream:

• Python 3.6
  
  $ pip install numpy pandas argparse datetime pafy streamlink flask_opencv_streamer youtube-dl flask-opencv-streamer opencv-contrib-python
• OpenCV with YOLOv3 detectionmethod (https://pjreddie.com/darknet/yolo/)
• Streamlink (https://github.com/streamlink/streamlink)
  
  
  

This repository can be use to do object detection in a video stream, then writing the the number of detected objects to an output file, every x seconds (default 5sec, dependent on performance). This is done by making use of the OpenCV library with the YOLOv3 detectionmethod. For an introduction to opencv and yolo refer to: https://www.pyimagesearch.com/2018/11/12/yolo-object-detection-with-opencv/. The sourcecode from this blog, by Adrian Rosebrock, is the starting point for this repository.

YOLO Weights:

Download the YOLOv3 weights:
$ wget https://pjreddie.com/media/files/yolov3.weights

The weights have been trained on the COCO dataset (http://cocodataset.org)

Example Output Jackson Hole Town Square Webcam

Example youtube streams: Jackson Hole Wyoming USA Live Cams - SeeJH.com

• https://youtu.be/1EiC9bvVGnk
• https://youtu.be/RZWzyQuFxgE
  
  
  
  Here I used:
• confidence level 0.30
• threshold level 0.55