Research On Image Enhancement And Night Vehicle Recognition Algorithm

In recent years,convolutional neural networks have been continuously applied in the field of target detection,and the accuracy of the detection system of vehicle detection systems has continued to refresh the upper limit.Even so,when the use scenario is converted to the night,the accuracy of the detection is no longer ideal,and the existing neural network detection method has certain limitations for vehicle detection.In this context,this thesis uses a combination of image enhanced technology and convolutional neural networks to improve the accuracy of night vehicle detection.This method is enhanced by night road images,and vehicles are detected in combination with convolutional neural networks to achieve accurate identification of night vehicles.The main research content and technical path are as follows:1.In-depth research on several image enhancement algorithms with high frequency use and propose image enhancement algorithms in this article on this basis.The theoretical research and experimental analysis of the histogram balanced,homomorphic filtering and Retinex algorithm,analyze the characteristics and advantages and disadvantages of the algorithm.The multi-scale Retinex algorithm with color recovery is more significant than other algorithms in the night image enhancement,but there are still some shortcomings.The algorithm performance still has room for improvement.2.Optimize the multi-pair Retinex algorithm algorithm with color recovery.While the improved image enhancement algorithm maintains the original color,the color ratio is relatively balanced,the detail information of the edge of the vehicle is completely retained,the outline of the object is relatively clear,the brightness is bright,the brightness is bright,the brightness is bright,and moderate contrast.3.Based on the YOLOV4 target detection network and image enhancement algorithm,the night vehicle detection system is set up.A two-point improvement strategy is proposed to the vehicle detection system: modifying the loss function and increasing the small target detection layer,and the effectiveness of the improvement strategy from the aspects of visualized experiment results and theoretical experimental data.The newly increased detection layer increased by 2.06% for the average detection accuracy of the vehicle detection network.The optimized loss function brought an increase of 0.98%to the average detection accuracy.This thesis studies the key technologies of vehicle detection in the night road environment,that is,image enhancement and vehicle detection algorithms,proposed a detection strategy that combines image enhancement algorithms and convolutional neural networks,and achieved good results.