Object Detection System Based On Vehicle-borne Infrared Image

Moving object detection in road traffic is an important part of driving assistant system.In recent years,most scholars have adopted the deep learning method to detect objects in visible images.While the ordinary vehicle-borne visible camera is vulnerable to darkness,strong light and bad weather,the use of infrared camera can greatly alleviate the above problems.However,object detection algorithm with infrared images in practical application has some limitations in robustness and accuracy.In this paper,moving object detection in infrared images under complex traffic scenarios is studied,details are as follows:(1)According to the gray statistics,noise and frequency characteristics of infrared images,preprocessing methods of image denoising and enhancement are studied in this paper.In image denoising,a comprehensive filtering method is established to effectively suppress noise in the image.In image enhancement,a bilateral filtering algorithm based on histogram processing is proposed,which combines the spatial information and statistical information of the image.The proposed algorithm can effectively maintain the details of the infrared image as well as enhance the contrast.Experimental results show that using the pre-processed infrared images can reduce the missed objects.(2)In object detection,the regression-based object detection algorithm is adopted.Based on YOLOv2(You Only Look Once 9000),the activation function of the network is modified and cluster analysis on the bounding boxes of the objects is carried out by means of dimension clustering.The above operations can not only ensure the robustness of the algorithm,but also effectively reduce the missed objects and false objects In order to verify the performance of the improved algorithm,training and testing experiments are conducted on public datasets in this paper.Experimental results show that the proposed algorithm has high accuracy and practicability in different complex road scenarios.In summary,problems in moving object detection in complex traffic environment based on infrared images are studied in this paper,which provides a new idea for the practical application of driving assistant system.48 figures,11 tables,50 references.