Research On A Traffic Video Monitoring Method Based On Visual Attention Mechanism

With the improvement of transportation infrastructure and economic development,the number of vehicles and drivers has soared,and public transportation safety problems have frequently occurred.The traditional intelligent traffic monitoring system is difficult to implement full coverage because it requires a lot of manpower and is difficult to operate.Therefore,it is necessary to use artificial intelligence to realize a new generation of intelligent traffic monitoring system.A visual simulation algorithm based on the attention model(referred to as the visual attention algorithm)can mark the more important areas in the traffic video for viewing,and can set up a training set to determine which aspect of the important area(such as human Or car).At the same time,it can be used as a preprocessing module to add other computer vision algorithms(such as target recognition and instance segmentation)and enhance the performance of the algorithm,so it has attracted widespread attention from researchers.Aiming at the shortcomings of the existing visual simulation algorithms,a traffic video surveillance method based on visual attention mechanism was proposed.The algorithm is improved from three aspects: public feature extraction module,attention module,and network lightweight.The public feature extraction module uses the residual network as the core idea,which deepens the network and enhances the ability to extract image features while avoiding network degradation.The attention module uses explicit attention and implicit attention to enhance the data from a spatial perspective and a channel perspective,respectively,to enhance the effect of the network.Use deep separable convolutions and 1 * 1 convolutions instead of standard convolutions to reduce network parameters.According to the actual use of the algorithm,new evaluation indexes THPS and PHPS are proposed,and the results obtained by the algorithm are combined with the original image to form an intuitive effect map for comparison.The experimental results show that the improved algorithm of feature extraction and attention module is basically the same as the traditional algorithm in THPS index,but it is improved by 25% on PHPS.From the visual effect point of view,the improved algorithm is even better.Although the performance of the lightweight network is reduced by 6%-7%,the parameters are only 3/10,and the operating speed and network training speed are both increased by about 20%,which meets the actual use requirements.The research on the visual attention mechanism algorithm for traffic monitoring not only paves the way for future use of computer vision algorithms such as target recognition and target segmentation on traffic monitoring images,but also optimizes it for mobile terminals,making the algorithm more applicable.