AI-Based Uncertainty-Aware Cooperative Perception using Dempster-Shafer Theory for Roadside-Vehicle Sensor Fusion

09.03.2026, Abschlussarbeiten, Bachelor- und Masterarbeiten

Develop AI-based cooperative perception that fuses roadside and vehicle sensor data using uncertainty modeling and Dempster-Shafer Theory to detect conflicts, handle occlusions, and improve multi-sensor object tracking reliability.

Motivation & Relevance

Roadside perception systems and connected vehicles generate complementary observations of the same traffic environment. Fusion of these sources can improve object detection robustness, especially under occlusion, adverse weather, or sensor noise. However, inconsistent or conflicting detections can degrade performance. Probabilistic uncertainty modeling combined with Dempster-Shafer Theory (DST) offers a principled way to combine uncertain information and manage conflicts between sensors. This enables more reliable cooperative perception for automated driving and intelligent infrastructure.

Project Description

In this thesis, you will implement a cooperative perception framework that fuses roadside cameras and vehicle sensors:

• Model epistemic and aleatoric uncertainties in object detections
• Implement Dempster-Shafer-based evidence fusion across sensors
• Detect and handle conflicts in object states
• Evaluate robustness under occlusion, false positives, and noisy detections
• Compare DST-based fusion against classical probabilistic fusion
  

The result will be a modular, uncertainty-aware multi-sensor fusion pipeline that can be applied to real roadside-vehicle scenarios.

Your Tasks

• Design cooperative perception fusion architecture
• Implement uncertainty modeling for roadside and vehicle detections
• Apply Dempster-Shafer Theory for conflict-aware fusion
• Conduct robustness evaluation and performance benchmarking
• Compare with baseline probabilistic fusion

Your Profile

• Master’s student in CS, Robotics, EE, or related field
• Strong Python/C++ programming skills
• Interest in AI-based perception and sensor fusion
• Knowledge of object detection, tracking, or uncertainty modeling is a plus

What you will gain

• Deep understanding of uncertainty-aware multi-sensor fusion
• Practical experience with cooperative perception for automated driving
• Exposure to evidential reasoning techniques (DST)
• Hands-on work with real AND simulated roadside/vehicle data

I am looking forward to receiving your application. Please include a very brief motivational statement and your grade transcript.

Kontakt: erik-leo.hass@tum.de