REAL-TIME FOREGROUND SEGMENTATION FOR SURVEILLANCE APPLICATIONS IN NRCS LIDAR SEQUENCES

Abstract. In this paper, we propose a point-level foreground-background separation technique for the segmentation of measurement sequences Non-repetitive Circular Scanning (NRCS) Lidar sensor, which is used as 3D surveillance camera mounted in fixed position. We show that by applying NRCS technology, can overcome various limitations rotating multi-beam sensors, such low vertical resolution, disadvantageous applications. As main challenge, need to efficiently balance between spatial and temporal resolution recorded range data. For reason, automatically generate maintain very high-resolution background model sensor’s Field View, while enabling real-time analysis dynamic objects use integration time extract consecutive frames. result, laser reflections from foreground reflect sparse, but geometrically accurate samples silhouettes providing valuable input higher-level shape description or event steps. demonstrate efficiency new approach different realistic measurements sequences, obtaining 0.76 overall F1-score on measured dataset.