Collision Detection: An Improved Deep Learning Approach Using SENet and ResNext

TL;DR

This paper proposes a deep learning collision detection model combining ResNext and SENet, outperforming existing models with less training data, aiming to enhance automotive safety systems.

Contribution

Introduces a novel ResNext-SENet based model that improves collision detection accuracy while reducing training data requirements.

Findings

Achieves ROC-AUC of 0.91 on collision detection task.

Outperforms VGG16, VGG19, Resnet50, and ResNext baseline models.

Requires less synthetic data for training.

Abstract

In recent days, with increased population and traffic on roadways, vehicle collision is one of the leading causes of death worldwide. The automotive industry is motivated on developing techniques to use sensors and advancements in the field of computer vision to build collision detection and collision prevention systems to assist drivers. In this article, a deep-learning-based model comprising of ResNext architecture with SENet blocks is proposed. The performance of the model is compared to popular deep learning models like VGG16, VGG19, Resnet50, and stand-alone ResNext. The proposed model outperforms the existing baseline models achieving a ROC-AUC of 0.91 using a significantly less proportion of the GTACrash synthetic data for training, thus reducing the computational overhead.