A Radar and Sensor-Aware Kalman Filter Approach for Advanced Driver Assistance Systems

Advanced driver assistance systems (ADAS) rely heavily on robust object tracking to ensure safe and autonomous navigation, especially in complex outdoor environments.Traditional Kalman filter (KF)-based methods, while effective in ideal conditions, often fall short in scenarios with high noise, asynchronous sensor data, occlusions, and varying environmental conditions.The existing tracking techniques do not adequately address the challenges of multi-object tracking under low Signal-to-Noise Ratio (SNR) or nonlinear dynamics.To bridge this gap, this work proposes Radar and Sensor-Based Tracking with Adaptive Spatial-Temporal Analysis (RASTA), a modified KF-based architecture designed to enhance multi-object tracking using mmWave radar in ADAS.The primary objective of this work was to improve tracking accuracy, handle sensor uncertainty, and enable robust performance in dynamic and noisy conditions.The methodology involved simulating ADAS motion using a discrete Langevin process with bistable dynamics, converting Cartesian trajectories to polar coordinates, and introducing noise to emulate real-world radar behavior.Experimental validation using a mmWave dataset showed that RASTA achieved up to 12.4% improvement in azimuth estimation and 10.7% in radial distance accuracy over baseline methods.The results show RASTA's effectiveness in delivering reliable, accurate tracking.