Indoor Temporally Constrained Instantaneous Ego-Motion Estimation Using 4-D Doppler Radar

Indoor ego-motion estimation using millimeter-wave Doppler sensors is challenging due to high levels of outliers, primarily caused by multipath reflections. A standard approach to mitigate these outliers is random sample consensus (RANSAC), where the ego-motion model is derived from data collected at a single time step, overlooking the continuity between successive measurements. In this article, we demonstrate that leveraging temporal relationships across multiple time steps can improve ego-motion estimation accuracy in indoor environments. We introduce two novel RANSAC-based methods that incorporate a weighted sliding window to enhance ego-motion estimation: temporal sample consensus (TEMPSAC) and temporally weighted least squares (TWLSQ). In TEMPSAC, samples are selected with a probability-weighted by their temporal proximity, and the velocity model is generated using least-squares regression (LSQ). In TWLSQ, samples are uniformly selected, but the velocity model is parameterized with a temporally weighted LSQ. Both methods calculate the platform’s motion by prioritizing temporally consistent inliers. Experimental validation of 18 indoor trajectories shows an average position accuracy improvement of 27% compared to previous RANSAC-based ego-motion implementations. The results demonstrate the effectiveness of incorporating temporal information into mmWave-based ego-motion estimation.