An Attention-based Recurrent Convolutional Network for Vehicle Taillight\n Recognition

2019/06/09 by Kuan‐Hui Lee, Takaaki Tagawa, Lee, Kuan-Hui +7
Engineering · Computer Science · #Autonomous Vehicle Technology and Safety #Advanced Neural Network Applications #Video Surveillance and Tracking Methods

paper · pdf · doi:10.48550/arxiv.1906.03683

Abstract

Vehicle taillight recognition is an important application for automated\ndriving, especially for intent prediction of ado vehicles and trajectory\nplanning of the ego vehicle. In this work, we propose an end-to-end deep\nlearning framework to recognize taillights, i.e. rear turn and brake signals,\nfrom a sequence of images. The proposed method starts with a Convolutional\nNeural Network (CNN) to extract spatial features, and then applies a Long\nShort-Term Memory network (LSTM) to learn temporal dependencies. Furthermore,\nwe integrate attention models in both spatial and temporal domains, where the\nattention models learn to selectively focus on both spatial and temporal\nfeatures. Our method is able to outperform the state of the art in terms of\naccuracy on the UC Merced Vehicle Rear Signal Dataset, demonstrating the\neffectiveness of attention models for vehicle taillight recognition.\n

Related

• Traffic Light Recognition using Convolutional Neural Networks: A Survey
• TLD: A Vehicle Tail Light signal Dataset and Benchmark
• Robust Detection, Association, and Localization of Vehicle Lights: A Context-Based Cascaded CNN Approach and Evaluations
• Lane Attention: Predicting Vehicles' Moving Trajectories by Learning Their Attention over Lanes
• Cross-Domain Car Detection Model with Integrated Convolutional Block Attention Mechanism