Skip to content

Pedestrian tracking using Yolov4-tiny custom model for detection in OpenCV

License

Notifications You must be signed in to change notification settings

dipzza/pedestrian-tracking

Repository files navigation

pedestrian-tracking

Track pedestrians from video or camera using OpenCV.

It detects people bounding boxes using a custom yolov4-tiny model and tracks them.

Instructions

Person Detector

The initial model was the first 29 layers of Yolov4-tiny pre-trained on the COCO dataset on 80 classes.

Then it was trained / fine-tuned just to detect people with a 416x416 resolution for 6000 iterations on the CrowdHuman dataset improving the mean Average Precision (mAP) from 34.82% to 52.41%, and further to 61% increasing the resolution to 608x608 for inference.

Tracker

Thanks

To every contributor / researcher that has made possible the following:

About

Pedestrian tracking using Yolov4-tiny custom model for detection in OpenCV

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published