Research On Vehicle Assisted Driving System Based On Infrared Thermal Imaging Technology

As industrialization continues to progress,people’s living standards are improving,and cars have become a necessary means of travel,but at the same time,traffic accidents are gradually increasing.Especially when driving at night on rural roads,it is very easy to cause traffic accidents due to insufficient light and other reasons.The subject is aimed at the problem of nighttime vehicle driving on rural roads,based on infrared thermal imaging technology,and the use of image processing and pedestrian detection technology,to design a vehicle-assisted driving system capable of preliminary nighttime pedestrian recognition,to help drivers detect nighttime pedestrian targets and reduce the incidence of accidents.The research of this topic is as follows.Firstly,for the problem of noise in the collected infrared images,an improved hybrid filtering algorithm is proposed by studying the median filtering and mean filtering algorithms.By setting the upper and lower threshold limits,the pixel points larger than the upper threshold limit and smaller than the lower threshold limit are subjected to median filtering for eliminating pepper noise,and the pixel points between the upper and lower threshold limits are subjected to mean filtering for handling Gaussian noise.Next,an improved box plot-based adaptive segmented nonlinear transform image enhancement algorithm is proposed.The algorithm achieves the function of adaptive selection of segmentation parameters by sorting and partitioning the grayscale image pixel gray values to find the grayscale intervals with concentrated distribution.A point outside the right side of the grayscale concentrated distribution interval is selected as a segmentation point,and three times spline interpolation is performed on all segmentation points to generate a smooth nonlinear curve,and the nonlinear curve function is used as the transform function of the grayscale transformation of the infrared image to realize the nonlinear grayscale transformation.Again,an infrared pedestrian target detection method combining particle swarm optimization SVM plus NMS optimization is used.First,the acquired infrared pedestrian images are fed into a dataset of positive and negative samples,then the HOG features are extracted from the dataset,then the extracted features are fed into a particle swarm optimized SVM for training and difficult sample detection,and the NMS optimized detection window is obtained after the classifier,and the video sequences are optimized for limited region detection and interframe detection.Finally,the infrared thermal imaging-based in-vehicle assisted driving system is ported cross-platform to an in-vehicle host computer with Ubuntu 16.04 operating system to realize the in-vehicle assisted recognition function.In this paper,we use infrared images captured by infrared cameras as research materials,and perform image filtering,image enhancement and infrared pedestrian recognition for these infrared materials.Three improvements are proposed according to the studied content.First,for the captured infrared images are noisy,an improved hybrid filtering algorithm is proposed to realize the image noise reduction function in the early stage;second,for the image pedestrian target area is not prominent,an adaptive segmentation nonlinear algorithm based on box plot is proposed to realize the automatic determination of the segmentation point function and nonlinear transformation function in the grayscale concentration area,and the experimental results show that The proposed image enhancement algorithm can improve the brightness of the foreground target of the image,suppress the background color,make the image transition smooth,and achieve the adaptive nonlinear transformation and obtain good enhancement effect compared with the traditional algorithm;Third,for the uncertainty of SVM parameters in pedestrian recognition,the method of particle swarm optimization SVM is used,and through the NMS method,the overlapping target detection frame is removed to increase the detection accuracy,and through Simulation and experiments are conducted to get better results.