Monday, January 12, 2026

Deep Learning–Based Real-Time Detection of Vehicle Light Signals for Motion Prediction

Topic and Goal of the Thesis

Automated vehicles require reliable perception signals to anticipate the behavior of other road users at an early stage. Turn indicators and brake lights represent important visual cues for upcoming maneuvers.

The objective of this thesis is to train a real-time-capable deep learning model that uses a multi-camera system to detect the light states of surrounding vehicles and track them over time. Finally, the detected light states are associated with a provided external list of traffic participants.

Working Points

• Literature review on vehicle light / turn signal detection and detector–tracker architectures
• Development and training of a real-time–capable deep learning model
• Integration of a tracking and association algorithm for assigning light states to traffic participants
• Experimental evaluation of the complete pipeline with respect to detection accuracy, temporal stability, and real-time performance

Requirements

• Reliability, commitment and enjoyment of working independently
• Experience with Python
• Experience with the following is beneficial: Machine Learning, ROS and Docker

Note: Please attach brief resume and grade summary.