An ADAS pipeline that consists of Traffic Light, Traffic Sign detection and anterior car distance estimation is developed in this project. A deep learning approach is used to develop a robust object detection model. Different algorithms and deep learning frameworks are experimented with, to provide a detailed analysis as to which model performs the best for the particular task at hand.

The repo consists of scripts that help visualize detections from models that are trained jointly on different datasets - the Bosch Small Traffic Lights Dataset and the Tsinghua-Tencent 100K Traffic Sign dataset. The model has been tested on video sequences and achieves a good frame rate with detections from 50 classes which consist of traffic lights, traffic signs and cars. More details about the project can be found in the project_report.pdf file in the materials folder.

This project is based on TensorFlow's Object Detection API. However, note that the project was originally implemented using TensorFlow 1.14 and not TensorFlow 2.x