PillarFlow: End-to-end Birds-eye-view Flow Estimation for Autonomous Driving

TL;DR

This paper introduces PillarFlow, an end-to-end deep learning framework that estimates bird's eye view flow from LIDAR data, enhancing obstacle tracking for autonomous driving.

Contribution

It presents a novel LIDAR-based BeV flow estimation method that improves obstacle tracking accuracy in autonomous driving scenarios.

Findings

Accurately estimates 2-D BeV flow from LIDAR data.

Enhances tracking performance of dynamic and static objects.

Demonstrates superior accuracy over existing methods.

Abstract

In autonomous driving, accurately estimating the state of surrounding obstacles is critical for safe and robust path planning. However, this perception task is difficult, particularly for generic obstacles/objects, due to appearance and occlusion changes. To tackle this problem, we propose an end-to-end deep learning framework for LIDAR-based flow estimation in bird's eye view (BeV). Our method takes consecutive point cloud pairs as input and produces a 2-D BeV flow grid describing the dynamic state of each cell. The experimental results show that the proposed method not only estimates 2-D BeV flow accurately but also improves tracking performance of both dynamic and static objects.