This is done using YOLO (You Only Look Once), a Convolutional Neural Network (CNN) that has been trained to detect a variety of different objects. This project in particular uses YOLO v3, which can detect 80 different kinds of objects!
To persist object ID’s, a customized version of sort.py is used. Our contributions to this allow us to also include object labels with each tracker, and retrieve the dead tracklets for submission to our server.
Results of tracking, categorized by Type. You can see signs of jay walking, but stop patterns, and more.
Results of tracking, categorized by Direction. You can see traffic flow and problematic driving habits (such as drivers drifting into bike or turn lane)