Research On Pedestrian Detection Algorithm In Infrared Images

Infrared images are obtained by the infrared sensor with its thermal imaging performance, which are only dependent on temperature and heat radiation of the objects. Therefore, compared to the visible images, infrared images have an obvious advantage at night or under the condition of insufficient light. The technologies of pedestrian detection and tracking based on infrared have played an important role in night intelligent surveillance, vehicle safety driving and other areas. Detecting and tracking pedestrian target is a hot topic and challenging research in object detection and tracking field as human target is the most important and most active factor in environment. However, this task is full of difficulties and challenges because of the none-rigid human body and the infrared image’s shortcomings.In this thesis, the static pedestrian targets in one frame infrared image are treated as the research object. In order to improve pedestrian detection rate, conducted research on related algorithms. The main content of the thesis include the following:(1) Introduced the principle of infrared imaging image, compared infrared images with visible images. Difficulties and problems of pedestrian detection in infrared images were summarized. Then, classified and compared pedestrian detection methods in infrared images, analyzed the advantages and disadvantages of different algorithms. Finally, several common pedestrian feature descriptors and machine learning methods were highlighted.(2)In order to solve the problems of the low detection rate of single feature descripting pedestrian insufficiently. An infrared pedestrian detection algorithm based on intensity self-similarity and histogram of oriented gradients was introduced in this thesis. It combined the contours of pedestrian and local similarity information. In order to achieve the purpose of pedestrian detection, support vector machine was used to learn and classify. Experimental results demonstrated the effectiveness of the algorithm on LSI Far Infrared Pedestrian Dataset. Compared with method based on single feature, the detection rate of method based on multi-features was significantly better than the method based on single feature.