Traffic accident detection is a challenging task in intelligent transportation systems. Rapid and accurate detection of traffic accidents can ensure the safety of people's lives and property to the greatest extent. However, the existing accident detection models have some limitations such as slow speed or low precision, which limit the practical application. Aiming at the current low accuracy of rapid traffic accident detection, this paper proposes a traffic accident detection algorithm based on attentional feature fusion, which combines the feature extraction and fusion process with the attention mechanism. Specifically, the MHSA module is integrated into the Backbone stage of the YOLOv5 network, and the CBAM module is integrated into the Neck stage to refine and capture image features and enhance network feature extraction and fusion capabilities. At the same time, in view of the lack of traffic accident image data, this paper establishes a traffic accident image data set. Experiments were carried out on the traffic accident image data set and the results show that, under the condition that the detection speed of each image is guaranteed to be 9.7ms, the accuracy of the proposed algorithm to identify traffic accidents is 4% higher than that of YOLOv5s. It achieves a balance between speed and accuracy of traffic accident detection under limited computational cost.
Lightweight Traffic Accident Detection Algorithm Based on Attention Mechanism
29.03.2024
582186 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Traffic Accident Anticipation via Driver Attention Auxiliary
Springer Verlag | 2024
|A Lightweight Uav Swarm Detection Method Integrated Attention Mechanism
DOAJ | 2022
|Traffic accident detection device and method of detecting traffic accident
Europäisches Patentamt | 2017
|