Real-Time 2D LiDAR Object Detection Using Three-Frame RGB Scan Encoding

TL;DR

Problem: privacy-friendly indoor perception with 2D LiDAR on embedded hardware. Approach: encode short-term temporal context by stacking three consecutive scans as RGB channels to form a $64\times384\times3$ input for YOLOv8n, avoiding occupancy-grid preprocessing. Contributions: (i) compact three-frame RGB LiDAR encoding, (ii) sim-based evaluation with holdout yielding mAP@0.5=$0.984$, mAP@0.5:0.95=$0.778$, and (iii) real-time embedded inference on Raspberry Pi 5 with latency $=$ $47.8$ ms, plus a lower-end occupancy-grid baseline comparison. Significance: demonstrates that lightweight temporal encoding enables accurate, real-time LiDAR-only detection in privacy-sensitive indoor robotics and supports sim-to-real exploration.

Abstract

Indoor service robots need perception that is robust, more privacy-friendly than RGB video, and feasible on embedded hardware. We present a camera-free 2D LiDAR object detection pipeline that encodes short-term temporal context by stacking three consecutive scans as RGB channels, yielding a compact YOLOv8n input without occupancy-grid construction while preserving angular structure and motion cues. Evaluated in Webots across 160 randomized indoor scenarios with strict scenario-level holdout, the method achieves 98.4% mAP@0.5 (0.778 mAP@0.5:0.95) with 94.9% precision and 94.7% recall on four object classes. On a Raspberry Pi 5, it runs in real time with a mean post-warm-up end-to-end latency of 47.8ms per frame, including scan encoding and postprocessing. Relative to a closely related occupancy-grid LiDAR-YOLO pipeline reported on the same platform, the proposed representation is associated with substantially lower reported end-to-end latency. Although results are simulation-based, they suggest that lightweight temporal encoding can enable accurate and real-time LiDAR-only detection for embedded indoor robotics without capturing RGB appearance.

Figures (6)

• Figure 1: RGB-encoded LiDAR input: three consecutive scans are rasterized and stacked (red: $t{-}2$, green: $t{-}1$, blue: $t$) to form a compact tensor used by YOLOv8n.
• Figure 2: Visualization of an explored "aligned fused" encoding using a 5-frame buffer (hit counts, edges, and density). This was not used in the final pipeline due to added cost without clear accuracy gains.
• Figure 3: Example randomized scenarios with 90 predefined robot positions. Object types are shown as colored rectangles (box: blue, chair: green, desk: purple, doorframe: red) and robot waypoints as blue dots.
• Figure 4: 3D view of the Webots simulation environment showing a typical indoor scenario with objects (box, table, chair, doorframe) and the room structure.
• Figure 5: Example RGB-fused LiDAR input with automatically generated bounding boxes for the four object classes.