Danger Monitoring for Cyclists using Raspberry Pi and Object Detection

Herberto Warner proposes a low-cost danger monitoring system for cyclists using computer vision and object detection to enhance safety on the streets. The prototype involves a Raspberry Pi 4, a camera, and an LED traffic light system that communicates danger levels to cyclists by detecting cars and bicycles. The project utilized three datasets, including real-life images captured from a cyclist's perspective and pre-labeled images from Google Open Images, to train the system. The data was preprocessed using Roboflow, which facilitated the creation of TFRecords and accelerated the development process. Employing the MobileNetSSD model for its fast inference time, the system evaluates danger based on object distance and size, with the LED light signaling different danger levels. The model excelled in detecting larger objects like cars but struggled with smaller objects such as bicycles, prompting future improvements in data collection and model exploration. Warner envisions the system's potential integration into smart cities to promote safer urban cycling environments.