SUMMARY

This thesis discusses the implementation of a traffic light learning & prediction model. The increase in V2X communications to predict traffic light behavior live, efforts to improve fuel economy, and desires to cut environmental pollution due to vehicle emissions are the main motivations for this project. Vehicles with a camera use a vision system to detect and upload signal states into a central learning database. A batch updating procedure runs on this data to develop/refine signal length predictions and stores them in a knowledgebase. Traffic lights are detected with greater than 90% success rate & drivers are always informed on the upcoming signal and its predicted change time at a good distance from the light. Predictions are determined with a high probability of capturing signal drift and changes in light schedule.