Research On Visual Image Enhancement Method For Unmanned Vehicle Object Detection

The accuracy of object detection plays a vital role in the safety and intelligence of unmanned vehicles.Among them,the accuracy of unmanned vehicle object detection in harsh environments is mainly guaranteed by image enhancement of degraded images.In this paper,the research on image enhancement methods for unmanned object object detection under harsh environments such as low illumination and foggy weather is carried out.The main research contents are as follows:1.The research on the difference enhancement method of low-light RGB image based on the improved Retinex algorithm is carried out in this paper.In this research,spatial conversion is combined to improve the running speed,the logarithmic function is replaced in the multiscale Retinex algorithm by the sigmoid function to broaden the image gray range,the Gaussian filter is replaced by the edge-preserving bilateral filtering as the surround function to protect the edge contour of the image,the color restoration function is introduced to restore the color information low-light image.A object detection algorithm based on YOLO v3 is used to compare the enhanced images with the original low-light images.The experiment shows that the quality of the enhanced image and the accuracy of object detection have been greatly improved2.The method of frequency division enhancement for foggy RGB image based on improved homomorphic filtering is developed in this paper.This method integrates the HSI color space to increase the processing speed of the algorithm and at the same time improve the color fidelity of the image.It integrates homomorphic filtering to ensure the completeness of image details and contour information,histogram equalization to suppress noise.The YOLO v3-based object detection algorithm is exploited to compare the enhanced images with the original foggy images.The experiment shows that the quality of the enhanced images and the accuracy of object detection have been greatly improved.3.Aiming at the problem of incomplete defogging in the improved foggy RGB image frequency division enhancement method in dense fog weather,the improved foggy RGB image frequency division enhancement method combined with deep learning on the basis of this method is develpoed.The YOLO v3-based object detection algorithm is used to compare the enhanced images with the original foggy image.The experimental results show that the quality of the enhanced images and the accuracy of object detection have been greatly improved.4.In view of the cumbersome calling process of the enhancement algorithm and the unintuitive display of the contrast before and after enhancement,the research on an image adaptive enhancement and object detection system is carried out.And a new gray threshold parameter is introduced to perform adaptive enhancement and object detection on the images,to facilitate the enhancement and detection operations of different types of images.Meantime,the system can display the performance parameters of the image.And besides,it realizes the intuitive contrast show of degraded images,enhanced images and detected images.