Low Cost ESP32 Vision-Based Edge AI Device for Bicycle Rear Vehicle Detection and Collision Warning

Overview

Cyclists face significant risk from rear-end collisions due to limited awareness of approaching vehicles, and existing safety technologies like the $200 Trek CarBack Radar remain pricey. This project investigated whether a low-cost ESP32-based edge AI system could reliably detect approaching vehicles and provide real-time alerts. The hypothesis was that embedded computer vision could achieve functional vehicle detection while staying below commercial costs. A rear-facing OV2640 camera with the ESP32 was mounted beneath a bicycle seat for continuous image capture and processing. Approximately 160 images were collected and labeled to train a MobileNetV2 object detection model optimized with Int8 quantization. The system was tested with controlled vehicle approach testing at environments, with LED alerts indicating safe, caution, and danger zones for the rider. Design iterations included model retraining, camera setting adjustments, and implementing a frame averaging algorithm. The final model achieved significantly higher accuracy than initial versions, with false positives reduced by 64% after frame averaging. Latency was consistent at 143 milliseconds per frame. Battery testing showed around 60-minute of continuous runtime. Dataset size, lighting conditions, and frame averaging drastically improved performance. The main error sources included lighting variability, motion blur, and unknown environments with new objects. Iterative design improvements proved critical for achieving practical reliability despite model limitations. The hypothesis was supported: embedded edge AI can provide effective rear-vehicle detection for cyclists at under $30. This validates camera-based AI vision as an accessible alternative to expensive technologies and suggests future applications in sensing and tracking.

Competition history

• CSEF 2026 Electronics & Electromagnetics (Senior Division) · Entry S-10-16

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Physics and Astronomy projects
• CSEF Physics and Astronomy projects
• Every CSEF 2026 project