02 / Congressional App Challenge

Safe Crossing

Congressional App Challenge2024Winner — CA District 17

Visually impaired pedestrians face life-threatening risks at crosswalks. Existing solutions rely on audible signals that aren't available at most intersections. A real-time system that detects vehicles, traffic lights, and crosswalk boundaries could provide the spatial awareness needed for safe crossing.

My Role

  • Trained YOLOv5 model for vehicle detection with custom crosswalk dataset
  • Built Keras classifier for traffic light state recognition
  • Developed crosswalk centering algorithm using computer vision boundary detection
  • Integrated multi-model pipeline with real-time audio feedback

Approach

We built a multi-model pipeline combining YOLOv5 for object detection, a Keras CNN for traffic light classification, and OpenCV for crosswalk boundary analysis. The system computes the user's offset from the crosswalk center (averaging 16.5ft drift in testing) and provides directional audio cues to keep them centered and safe.

PythonYOLOv5KerasOpenCVReact Native

Detection Pipeline

SAFE TO CROSSCLICKYOLOv5 — Car 96.1%YOLOv5 — Car 93.7%YOLOv5 — Car 97.4%

Click the traffic light to toggle red/green. Hover over cars for detection metadata.

Multi-Model Classification

Bounding boxes animate on scroll. Each detection demonstrates the multi-model pipeline: YOLOv5 + Keras + OpenCV.

Links

Next Project

Algae TrackerAlgae Tracker