Synthesizing and Identifying Noise Levels in Autonomous Vehicle Camera Radar Datasets

TL;DR

This paper introduces a synthetic data augmentation pipeline for autonomous vehicle datasets to simulate sensor failures and noise, and evaluates a neural network for noise recognition with moderate accuracy.

Contribution

It presents a novel synthetic data augmentation method for simulating sensor noise and failures in AV datasets, and provides a baseline neural network for noise detection.

Findings

Achieved 54.4% accuracy in noise recognition across 11 categories.

Created a realistic augmented dataset simulating sensor failures.

Demonstrated the effectiveness of the augmentation in training noise detection models.

Abstract

Detecting and tracking objects is a crucial component of any autonomous navigation method. For the past decades, object detection has yielded promising results using neural networks on various datasets. While many methods focus on performance metrics, few projects focus on improving the robustness of these detection and tracking pipelines, notably to sensor failures. In this paper we attempt to address this issue by creating a realistic synthetic data augmentation pipeline for camera-radar Autonomous Vehicle (AV) datasets. Our goal is to accurately simulate sensor failures and data deterioration due to real-world interferences. We also present our results of a baseline lightweight Noise Recognition neural network trained and tested on our augmented dataset, reaching an overall recognition accuracy of 54.4\% on 11 categories across 10086 images and 2145 radar point-clouds.