Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SSCATeR: Sparse Scatter-Based Convolution Algorithm with Temporal Data Recycling for Real-Time 3D Object Detection in LiDAR Point Clouds

Alexander Dow, Manduhu Manduhu, Matheus Santos, Ben Bartlett, Gerard Dooly, James Riordan

TL;DR

The paper tackles the latency challenge of real-time 3D object detection in LiDAR point clouds for drone-based SAD. It introduces SSCATeR, a temporal data recycling scheme that reuses convolution results in unchanged regions by employing a sliding 100 ms time window with 10 ms strides and a change-map-driven backbone, integrated into PointPillars. Empirically, SSCATeR achieves up to 6.61× per-layer speedups and 3.8–4.1× backbone improvements across datasets and embedded hardware, while preserving identical outputs to baseline sparse convolutions. The approach enables real-time onboard detection for drone swarms and shows compatibility with multiple architectures, with crosstalk interference posing minimal risk under typical swarm configurations.

Abstract

This work leverages the continuous sweeping motion of LiDAR scanning to concentrate object detection efforts on specific regions that receive a change in point data from one frame to another. We achieve this by using a sliding time window with short strides and consider the temporal dimension by storing convolution results between passes. This allows us to ignore unchanged regions, significantly reducing the number of convolution operations per forward pass without sacrificing accuracy. This data reuse scheme introduces extreme sparsity to detection data. To exploit this sparsity, we extend our previous work on scatter-based convolutions to allow for data reuse, and as such propose Sparse Scatter-Based Convolution Algorithm with Temporal Data Recycling (SSCATeR). This operation treats incoming LiDAR data as a continuous stream and acts only on the changing parts of the point cloud. By doing so, we achieve the same results with as much as a 6.61-fold reduction in processing time. Our test results show that the feature maps output by our method are identical to those produced by traditional sparse convolution techniques, whilst greatly increasing the computational efficiency of the network.

SSCATeR: Sparse Scatter-Based Convolution Algorithm with Temporal Data Recycling for Real-Time 3D Object Detection in LiDAR Point Clouds

TL;DR

The paper tackles the latency challenge of real-time 3D object detection in LiDAR point clouds for drone-based SAD. It introduces SSCATeR, a temporal data recycling scheme that reuses convolution results in unchanged regions by employing a sliding 100 ms time window with 10 ms strides and a change-map-driven backbone, integrated into PointPillars. Empirically, SSCATeR achieves up to 6.61× per-layer speedups and 3.8–4.1× backbone improvements across datasets and embedded hardware, while preserving identical outputs to baseline sparse convolutions. The approach enables real-time onboard detection for drone swarms and shows compatibility with multiple architectures, with crosstalk interference posing minimal risk under typical swarm configurations.

Abstract

This work leverages the continuous sweeping motion of LiDAR scanning to concentrate object detection efforts on specific regions that receive a change in point data from one frame to another. We achieve this by using a sliding time window with short strides and consider the temporal dimension by storing convolution results between passes. This allows us to ignore unchanged regions, significantly reducing the number of convolution operations per forward pass without sacrificing accuracy. This data reuse scheme introduces extreme sparsity to detection data. To exploit this sparsity, we extend our previous work on scatter-based convolutions to allow for data reuse, and as such propose Sparse Scatter-Based Convolution Algorithm with Temporal Data Recycling (SSCATeR). This operation treats incoming LiDAR data as a continuous stream and acts only on the changing parts of the point cloud. By doing so, we achieve the same results with as much as a 6.61-fold reduction in processing time. Our test results show that the feature maps output by our method are identical to those produced by traditional sparse convolution techniques, whilst greatly increasing the computational efficiency of the network.

Paper Structure

This paper contains 23 sections, 2 equations, 28 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Point-Based Object Detection
  4. Sparse Convolutions
  5. LiDAR as a Streaming Sensor
  6. Methodology
  7. Dataset Collection
  8. PointPillars Architecture and Modifications
  9. Sparse Scatter-Based Convolutions with Temporal Data Recycling: SSCATeR
  10. Data Reuse in Convolution
  11. Comparison with a Sparse Backbone and OpenPCDet models
  12. PandaSet Testing
  13. LiDAR-to-LiDAR Crosstalk Interference Testing
  14. Results
  15. Analysis and Evaluation of Pillar Feature Net Modifications
  16. ...and 8 more sections

Figures (28)

  • Figure 1: Two DJI M300 RTK drones midflight (top left), with railway bridge (right). The left-most drone is equipped with a Zenmuse L1 LiDAR. The drone to its right is equipped with the Zenmuse P1 photogrammetry camera.
  • Figure 2: Non-Repetitive Scanning Pattern of the DJI Zenmuse L1 over $100$ milliseconds.
  • Figure 3: A point cloud over the over-water railway bridge collected in Newcastle, County Wicklow, Ireland.
  • Figure 4: Livox Avia mounted below a DJI M300 RTK drone, with embedded computer system mounted on top.
  • Figure 5: Close-range photogrammetry of the railway bridge.
  • ...and 23 more figures