Moving vehicle tracking and scene understanding: A hybrid approach

In this paper, we present a novel deep learning method for detecting and tracking vehicles within the context of autonomous driving, particularly focusing on scenarios related to vehicle failures. Ensuring the precise identification and monitoring of vehicles is paramount for enhancing road safety in autonomous driving systems. Our contribution involves the introduction of a hybrid Siamese network that merges the capabilities of YOLO models with Transformers. This integration aims to address the limitations of Convolutional Neural Networks (CNNs) in grasping high-level semantic nuances, thereby facilitating accurate detection and tracking of multiple vehicles within a given scene. Beyond this, we also curated the traffic scene dataset, which serves as a resource for training a multi-vehicle tracking model specifically tailored to the unique characteristics of traffic environment.