PMD Detection

The problem

Personal mobility device (PMD) adoption has recently surged in major cities in South East Asia. While many of the riders are responsible, some of them are reckless in riding the PMDs, often speeding along pavements, leading to avoidable deaths and injuries.

What we did

We worked on developing an end-to-end computer vision pipeline that attempts to identify PMD riders who are travelling higher than the allowed speed. We trained and used FasterRCNN and YOLO for identification of PMDs in a video frame. Inference in actual deployment was accelerated using Nvidia’s Jetson TX1 and Intel’s Movidius Stick.

Watch a video demonstration of the system here: