Design Of Pedestrian Detection System Based On Infrared Imaging

In the process of the driving,there are a lot of risk because it will be influenced by different factors,.In order to achieve the goal of improving traffic safety,especially for the situation of night driving,driving for a long time and so on.A superior performance of the auxiliary driving system can achieve the result which improves the safety of the car and reduces automobile all kinds of traffic accidents.The thesis introduces the research background and significance of pedestrian detection.Analyzed the research status at home and abroad,and analyzed the difficulties and key points of pedestrian detection based on infrared imaging.With the gradual and widespread application of neural networks in pedestrian detection,the paper also elaborated the basic theories and concepts in deep learning.Based on the problems in the traditional SSD pedestrian detection algorithm,this article mainly develops the following aspects of work and innovation:(1)A pedestrian detection algorithm based on a shallow feature fusion network is proposed:The paper proposes a shallow feature fusion network detection algorithm based on SSD network.The algorithm refers to the structure of the original SSD network,and the original 38*38 and 19*19 prior frame output layers are revised to the maximum value.At the same time,the revised output layer and the original output layer are fused to obtain a new Feature fusion layer,and use it as the output network layer of the prior frame,and finally return the output prior frame to the input image.The algorithm improves the accuracy rate on the same input image by 0.1-0.2.(2)Image preprocessing based on adaptive brightness adjustment:The thesis analyzes the difference between visible light imaging and infrared imaging in detail,and input the original image directly into the target detector used in the subject,the detection accuracy will be greatly affected.In response to this problem,the paper proposes a processing method based on adaptive brightness adjustment.By modifying and reconstructing the brightness features of the image,the detail features and texture features of the input image are more obvious.In addition,the thesis replaces the optical transformation of traditional SSD network with this kind of processing method,so that the detection accuracy is optimized...