Research On Driver Detection Method Based On Image Enhancement In Traffic Surveillance Environment

With the rapid development of road traffic,the incidence of traffic accidents is increasing year by year.In order to cope with the severe traffic safety situation,it is essential to realize high-precision detection of drivers.Therefore,this article studies the driver detection method based on image enhancement in traffic surveillance scene.Starting from directly improving the accuracy of the detection network,this article studies a driver detection method based on improved Cascade R-CNN.On this basis,starting from the preprocessing of image enhancement,the driver detection method based on super-resolution and illumination enhancement is studied.The main contents of the article are as follows:1.Driver detection method based on improved Cascade R-CNN network.Aiming at the problem that Cascade R-CNN has low accuracy in detecting small-sized,occluded,and unobvious driver regions in the surveillance scene,this article proposes a driver detection algorithm based on the improved Cascade R-CNN network.The algorithm adds the FPN feature pyramid network module,replaces the Res Net-152 high-performance backbone network,and introduces the CBAM attention mechanism.These improvements can enhance the ability of algorithm to detect small-sized drivers and to distinguish driver characteristics.The experimental results verify that the driver detection algorithm designed in this article significantly improves the accuracy of driver detection in traffic surveillance scenarios.2.Driver detection method based on super-resolution processing.After analyzing difficult samples,it is found that the resolution of some surveillance images is too low,which makes it difficult to further improve the accuracy of driver detection by improving the detection network.Therefore,this article designs a super-resolution image enhancement algorithm based on improved EDSR network.Aiming at the problems that EDSR has poor ability to reconstruct image details and is sensitive to noise,this algorithm introduces a wider activation function block and pyramid attention mechanism,which improves the ability of algorithm to reconstruct image details and to suppress noise.This improves the image quality of the low-resolution driver area in the surveillance scene.The experimental results show that by combining the super-resolution algorithm with driver detection,the detection accuracy of drivers in traffic surveillance scenarios is improved.3.Driver detection method based on complex illumination enhancement.The application scenario of traffic surveillance is an outdoor environment,so the monitoring image is greatly affected by light,which makes it difficult to improve the accuracy of the detection network.In order to reduce the influence of illumination factors on the detection effect,this article designs a complex illumination image enhancement algorithm based on the improved Retinex-Net network.Aiming at the problems that Retinex-Net has inaccurate control of image information and generates more noise,the algorithm introduces the SENet attention module and U-Net++ network structure,which improves the attention of algorithm to image global information and suppresses noise.This improves the image quality of the driver’s area that is greatly affected by the illumination in the surveillance scene.The experimental results show that by combining the illumination enhancement algorithm with driver detection,the detection accuracy of drivers in traffic surveillance scenarios is improved.