visualbuffer/copilot

The system performs real-time lane and obstacle detection optimized for mobile deployment, using HLS color-space thresholding with adaptive luminosity normalization for lane boundaries and periodic YOLO inference (configurable intervals) to reduce computational overhead. Architecture includes a top-view perspective transformation for collision prediction and ego-vehicle localization, with tunable parameters for illumination robustness and lane-finding thresholds across diverse road conditions. Pre-trained Keras/YOLOv3 weights enable immediate video processing without training, making it deployable on resource-constrained mobile devices targeting ADAS applications in emerging markets.

177 stars. No commits in the last 6 months.