ScholarWorks@성균관대학교

초록

This paper proposes an efficient, real-time method for recognizing and tracking multiple objects using sparse point-cloud sequences generated by a low-cost mmWave radar. The system employs the DBSCAN algorithm to cluster the radar’s point cloud data, capturing objects across multiple frames within defined windows. A moving average filter is applied to mitigate measurement errors in the radar data. During the tracking phase, a Kalman filter predicts object positions, while the Hungarian algorithm ensures the correct assignment of detections to specific tracks. The proposed method is evaluated through five experiments, where people move within the radar’s field of view. These experiments involve overlapping people, making the tracking algorithm particularly challenging. The Multi-Object Tracking Accuracy (MOTA) metric is used to assess the results, achieving a 90.10% accuracy rate, which underscores the method’s potential for real-time multi-object tracking using mmWave radar. Videos of the experiments can be accessed via the following link: https://github.com/nabihandres/RADAR_tracking_tests.git © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

키워드