In autonomous driving, efficiently and accurately recognizing taillight states using dashcams is essential for interpreting other vehicles’ intentions. Recent video-based taillight recognition methods outperform earlier image-based approaches. However, they face challenges in efficiently integrating spatiotemporal information and managing high computational costs. In this paper, we introduce ETR, an accurate and efficient Transformer-based video classification model designed to enhance taillight recognition. Specifically, we first design a lightweight backbone to extract temporal and spatial features from videos and generate queries with prior information. Next, we develop a hierarchical progressive Transformer decoder that integrates feature maps from different levels of the backbone to enhance the model’s global information. Finally, we employ a classification head to predict the taillight state of the video. Additionally, we introduce a public dataset, ETR-Taillights, to address the current lack of open datasets for vehicle taillight recognition. The dataset contains 28,799 dashcam video clips, making it the largest public taillight recognition dataset. Experiments show that our method achieves a 91.69% F-measure on the ETR-Taillights dataset, surpassing the latest taillight recognition methods, VIF by 6.94% and CNN-LSTM by 10.82%. Additionally, it achieves an inference speed of 45.06 FPS, being 3.6 times faster than VIF. Furthermore, we conduct real-world road tests to demonstrate our method’s robustness and effectiveness in practical scenarios. Our model and dataset are available athttps://github.com/yang590/vehicle-taillight.

Vehicle light inspection is a crucial aspect of vehicle inspection, as vehicle lights play a vital role in understanding vehicle behavior. Due to diverse ambient illumination, non-uniform vehicle light standards, and complex light states, existing vehicle light recognition methods perform poorly, especially in turn signals. To realize intelligent vehicle light inspection, an accurate and real-time intelligent vehicle light recognition method is urgently needed. To address this issue, we propose a Transformer-based vehicle light recognition method, TrVLR, which effectively models sequential information and global relationships of the vehicle light image sequence, resulting in precise recognition. Specifically, we first utilize a convolutional neural network (CNN) backbone to extract features from vehicle light area video. Then, we adopt a Transformer encoder to encapsulate the long sequential information of the entire video. Next, a Transformer decoder is used to contextualize each image throughout the video and output the classification information of each image. Finally, we predict the state of the lights in each image by feed-forward networks (FFNs). Experiments demonstrate that our method obtains state-of-the-art results in headlights, taillights, and turn signals with F-measures of 93.34%, 91.64%, and 88.17%, respectively. Notably, TrVLR surpasses CNN-LSTM by 7.01% in turn signal recognition. Our method has been successfully applied to intelligent vehicle light inspection, and the video demo is shown athttps://www.youtube.com/watch?v=p8E78Pwkcxo.