An Empirical Analysis of Range for 3D Object Detection

TL;DR

This paper empirically analyzes the challenges of far-field 3D object detection in autonomous vehicles, revealing that different encoding strategies are needed for near- and far-field objects, and proposes ensemble methods to improve long-range detection.

Contribution

It introduces a detailed empirical analysis of far-field detection, and proposes range-specific encoding and ensemble techniques to enhance long-range 3D detection performance.

Findings

Near-field LiDAR data is dense and best encoded with small voxels.

Far-field LiDAR data is sparse and benefits from large voxel encoding.

Ensemble techniques improve long-range detection efficiency by 33% and accuracy by 3.2%.

Abstract

LiDAR-based 3D detection plays a vital role in autonomous navigation. Surprisingly, although autonomous vehicles (AVs) must detect both near-field objects (for collision avoidance) and far-field objects (for longer-term planning), contemporary benchmarks focus only on near-field 3D detection. However, AVs must detect far-field objects for safe navigation. In this paper, we present an empirical analysis of far-field 3D detection using the long-range detection dataset Argoverse 2.0 to better understand the problem, and share the following insight: near-field LiDAR measurements are dense and optimally encoded by small voxels, while far-field measurements are sparse and are better encoded with large voxels. We exploit this observation to build a collection of range experts tuned for near-vs-far field detection, and propose simple techniques to efficiently ensemble models for long-range detection that improve efficiency by 33% and boost accuracy by 3.2% CDS.