BrakeLightNet
Deep learning computer vision system for real-time brake-state understanding from dashcam footage. It detects vehicles ahead and classifies brake vs go states for driver-assistance and autonomous workflows. Achieved 92.44% validation accuracy, 84.5% recall, 93.4% precision, 0.887 F1-score, and ~30ms inference on Apple MPS with a lightweight 3.1M parameter model.