Research On The Algorithm Of Pedestrian Detection Based On Infrared Images

Pedestrian detection is one of the current research focuses in computer vision,which can also be widely adopted in intelligent video surveillance,vehicle assistant driving,population flow statistics and so on.The images captured by visible light cameras are always blurry when the light condition is poor at night or the weather is terrible,which makes pedestrian detection algorithm invalid based on visible light images.Due to the better transmission ability of the infrared radiation,less restricted by weather and illumination condition,humans can be captured by infrared cameras.Therefore,the research of infrared pedestrian detection is very important for practical applications.After analyzing the state of the art,this thesis proposes an improved infrared pedestrian detection algorithm,the main content of this thesis includes:(1)The thesis firstly outlines the technology of pedestrian detection based on infrared images.Then,the structure of the infrared pedestrian detection system and the problems existing in current infrared pedestrian detection algorithm are introduced.After analyzing the characteristics of the infrared images and the type of noise,the filtering method used in our work is described in this section.(2)Since the brightness of the pedestrian targets are not uniform,employing threshold segmentation method only often causes target fragmentation.For symmetrical pedestrians in infrared images,detection method based on vertical edge symmetry usually generates incorrect target locations.In order to solve these problems,an improved regions of interest extraction method has been proposed in this thesis.By the combination of adjacent frame subtraction and background subtraction as well as adaptive threshold segmentation technique based on statistical features,infrared video sequences and discrete infrared images are processed respectively to extract human candidate regions.(3)The adaptive histogram segmentation technique is employed to segment horizontally or vertically connected human candidate regions,when pedestrians are partly occluded by each other.According to the shape constraint of the pedestrian target,human candidate regions are refined by feature extraction method based on shape descriptors.Finally,pedestrians in infrared images are located after recognition process.(4)Infrared images captured by our infrared camera and the public infrared pedestrian database are used to test the proposed infrared pedestrian detection algorithm.The performance of the proposed method is analyzed by subjective evaluation and objective evaluation.By compared with other methods,the accuracy and robustness of the proposed method is proved.