Development of Novel Sensor Fusion Architectures for Autonomous Vehicles

Flux Auto mainly focuses on developing self-driving technology in trucks and cars to increase the viable commutability and sensor fusion plays an important role in providing a complete situational awareness for the vehicle even in harsh weather conditions like snow, rain, fog where different sensors work in different conditions.

Fusing the data from multiple sensors improves the 3D perception and determines the position, velocity, and orientation of other participants in the scenario. Different sensor fusion architectures like detection level fusion and track level fusion from Sensor Fusion and Tracking Toolbox™, Automated Driving Toolbox™ and MATLAB Coder™ were used to develop a novel architecture for multilevel fusion and was evaluated using OSPA and GOSPA metrics.

In this talk we will be talking about the following points:

• JPDA-based 3D multi-object tracking
• Density-based spatial clustering for detection-level fusion using GNN Tracker
• Implementation of ToMHT, GM-PHD, and JPDA for track-level fusion
• Effective filters and multi-object trackers for use on the sensors in autonomous stack
• C++ code generation for the sensor fusion architecture designed in MATLAB^® for real-time deployment and testing
• Multi-sensor data association for real-time and synthetic data
• Multi-sensor calibration in real time and concatenation of units