Research On Pedestrians Detection And Tracking Algorithm In Infrared Image Series

With the fast development of infrared technology, infrared devices have been widely used in more and more domains. Pedestrians are the main detection targets and objects in nighttime intelligent surveillance and control system and nighttime vehicle aided driving system. Pedestrian detection and tracking has also been receiving much concern in computer vision domain as a front direction and a hot topic. It possesses important theoretical meaning and extensive application value carrying out infrared pedestrian detection and tracking. The main contents researched are as below.The characteristics of infrared images are discussed firstly, then common moving objects detection methods are introduced and background generation methods are researched especially in this paper. For infrared images have low signal-to-noise ratio and contrast ratio, background image is generated using 3D median methods. The subtract-images are gained in background-subtraction method and enhanced subsequently. The main threshold segmentation methods are researched, and then, binary images are gained by segmenting subtract-images in Ostu methods, also be de-noised by morphological methods subsequently. Labeling the binary images, regions of interest are finally obtained.Through analyzing common description methods of objects'shape, principle and characteristics of Fourier Descriptor (FD) are researched. Subsequently, FD of ROIs in infrared images are extracted, so shape description of moving Objects are realized. Basing on BPNN, the pedestrian detection classifier is designed and used to identify ROI samples. The result shows that ,after being trained, the BPNN classifier designed in this paper has high true positive rate and low false positive rate, can determine whether one ROI is a person or not effectively.There are problems such as few target, much manual intervention, close distance, high resolution in common tracking methods. Corresponding relations of moving regions in successive frames are analyzed. One multi-pedestrian tracking algorithm based on moving regions is put forward. According to the test result by infrared image series, the method is robust and can track pedestrians adaptively and remotely.