Capturing Object Detection Uncertainty in Multi-Layer Grid Maps

TL;DR

This paper introduces a deep learning-based object detector for automated driving that estimates uncertainty in classification, pose, and shape, aiding trajectory planning with improved robustness and collision probability estimation.

Contribution

It presents a novel multi-layer grid map input and uncertainty estimation approach for object detection in autonomous driving, enhancing robustness and planning capabilities.

Findings

Models show comparable performance to existing detectors

Enhanced robustness in detection results

Effective approximation of collision probabilities

Abstract

We propose a deep convolutional object detector for automated driving applications that also estimates classification, pose and shape uncertainty of each detected object. The input consists of a multi-layer grid map which is well-suited for sensor fusion, free-space estimation and machine learning. Based on the estimated pose and shape uncertainty we approximate object hulls with bounded collision probability which we find helpful for subsequent trajectory planning tasks. We train our models based on the KITTI object detection data set. In a quantitative and qualitative evaluation some models show a similar performance and superior robustness compared to previously developed object detectors. However, our evaluation also points to undesired data set properties which should be addressed when training data-driven models or creating new data sets.