Study On Anti-halo Pedestrian Detection Technology Based On Heterogeneous Image Fusion In Night Vision

In the process of meeting cars at night,the driver’s sight is dazzled by the strong light of the opposite vehicle,and it is temporarily impossible to observe pedestrians and obstacles around the vehicle,which will induce serious traffic accidents and cause casualties.Therefore,anti-halo pedestrian detection appears particularly important.In order to achieve pedestrian detection when driving at night,this paper performs pedestrian detection based on heterogeneous source image fusion to eliminate haloing and improve driving safety at night.The following problems exist in the night vision anti-halo pedestrian detection technology: 1)The complexity of the fusion algorithm is high,resulting in low detection and tracking processing efficiency of the entire night vision anti-halation pedestrian detection system,and the speed is difficult to meet the actual requirements;2)Anti-halo light The fusion method cannot take into account image clarity,contrast and color richness while eliminating halo;3)The night vision anti-halo image is not as good as the daytime image quality due to the lack of light and poor imaging conditions at night.The difference in image quality between the fused night vision anti-halo image and the daytime image is large,resulting in low detection accuracy and poor tracking effect.To address the above problems the main research of this paper is as follows:1)In order to improve the real-time performance of the night vision anti-halation pedestrian detection and tracking system,an intermittent cyclic frame selection method based on the content difference between frames is proposed.The method reduces the number of fused video frames by comparing the content difference between two video frames and selecting the video frame whose content difference variation does not exceed but is closest to its threshold value as the fused frame.At the same time,the maximum frame interval threshold is introduced in the frame selection condition to avoid the problem that the video frame is not updated in time or the algorithm is wrong when the content difference between frames changes little.2)In order to solve the problem that it is difficult to take into account image clarity,contrast and color richness while eliminating halation in anti-halation images,an anti-halation method based on HSV color space and NSST is designed.This method decomposes the image into high and low frequencies through NSST transformation,adopts the fusion rule of automatic weight adjustment in the low frequency to achieve the purpose of eliminating halo,and a fusion rule based on regional contrast in the high frequencies to improve image contrast and sharpness,while combining HSV colour space to enrich the image colours.3)To improve the accuracy of pedestrian detection and tracking in anti-halo fusion images,a pedestrian detection algorithm P-YOLOv4 for night vision anti-halo background is designed,which enhances the pedestrian semantic features by adding cavity convolution in the backbone network;reduces the missed detection rate by learning the features of pedestrians of different sizes using the ASPP model;and improves the model by using k-means++ clustering a priori frame detection rate.To further ensure the accuracy of detection,the CL-KCF tracking algorithm is introduced to improve the accuracy of pedestrian detection and tracking in anti-halo images.The intermittent cyclic frame selection method based on inter-frame content differences proposed in this avoids the low real-time performance of the detection system caused by complex fusion algorithms;the designed anti-halo method based on HSV colour space and NSST both eliminates the high luminance information in the image and improves the richness of the image;the designed pedestrian detection and tracking methods of P-YOLOv4 and CL-KCF solve the problems of missed pedestrian detection and low tracking The problem of missed pedestrian detection and low accuracy is solved.