Cavon Hajimiri

A Wearable 10.5 GHz Radar-Based Pedestrian Collision Avoidance Sensor Using Edge-Executed Machine Learning

A car hits a pedestrian every 3 seconds and kills one every minute. Walkers and joggers face serious danger from vehicles, especially from behind, and currently no wearable solutions exist to protect them. To address this issue, I developed a wearable pedestrian radar, called Hindsight, that provides early warning alerts when vehicles are approaching from behind. The sensor detects cars from 100 meters away and gives the user vibrations and auditory alerts an average of 7 seconds before potential rear-end collisions. It uses a custom, ultra-low-cost Doppler radar and miniaturized machine learning model trained on a newly collected 170,000 frame dataset to detect approaching objects. Hindsight is small, affordable, and battery-efficient. The sensor operates for more than 20 hours at a time and keeps walkers and joggers safe by giving them extra time.