people-pathing detects and tracks paths of multiple people in a video. people-pathing leverages YOLO for object detection (detecting people in each frame) and SORT for object tracking (tracking people between frames). people-pathing generates plots of paths traversed by people in the video.
To use, ensure you have PyTorch and OpenCV installed in your environment.
The YOLO model requires the config directory in your project directory have the following files, which can be found via this link.
config
+-- coco.data
+-- coco.names
+-- yolov3.cfg
+-- yolov3.weights
Install the dependencies with the following command.
pip install -r requirements.txt
To run people-pathing, use the following command.
python app.py video_file_path [overlay_image_path] [-n=num_frames] [-s, --silent] [-d, --detection]
where
video_file_path is the path to the video file
overlay_image_path is the path to the image to overlay the measured paths on (optional, no overlay image by default)
-n=num_frames specifies the number of frames of the video to process (optional, all frames processed by default)
-s, --silent prevents STDOUT output from being displayed (optional, displays STDOUT output by default)
-d, --detection shows detection frame-by-frame while processing (optional, does not show frame-by-frame detection by default)
For example, the following command runs people-pathing on the video file data/videos/people_walking_mall.mp4 on the first 200 frames of the video with no STDOUT output displayed, and frame-by-frame detection shown while processing.
python app.py data/videos/people_walking_mall.mp4 -n=200 -s -d
The following has the same result but overlays the plotted paths on top of the image data/images/frame_1.jpg
python app.py data/videos/people_walking_mall.mp4 data/images/frame_1.jpg -n=200 -s -d
The following are plots generated by people-pathing with their corresponding commands.
Command: python app.py data/videos/people_walking_mall.mp4 -n=300 -s -d
Command: python app.py data/videos/people_walking_mall.mp4 data/images/frame_1.jpg -n=300 -s -d
- Rishi Masand
Object detection and tracking in PyTorch by Chris Fotache