Pedestrian Detection Under Low Resolution

Complex background is the key point of the current pedestrian detection research. Low resolution and low contrast limits the pedestrian detection accuracy. Studying these two problems will promote the pedestrian detection for engineering application. Pedestrian detection is largely applied to the video surveillance and the security industry as well as an important research field of the computer vision and the pattern recognition. To improve the academic significance and accelerate the pace of engineering application, this paper addresses the low resolution pedestrian detection.For static pictures of pedestrian detection, HAAR rectangle features of Viola-Jones are used to create appearance filters. Then design classifiers after AdaBoost training and generate static pedestrian detectors. For motion pedestrian detection, create motion pedestrian filters and get motion pedestrian detectors via training. Test data set is coming from PETS2001.For complex background problem, this paper analyzes the advantages and disadvantages of the Gaussian mixture model (GMM) based on moving objects detection method. To avoid GMM’s slow renewal rate and poor convergence, we are coming up with improvement methods. And the benchmark data set tests verify the effectiveness of the proposed method.The optical flow is used to match and track pedestrians. Then establish the pedestrian movement track for the pedestrian detection and classification. In order to eliminate nonhuman moving objects in the scene, this paper describes the shape information based method of pedestrian classification.According to the shadows of the complicated background, this paper describes the local binary pattern (LBP) texture operator and the shadow detection by HSV color space. And create the shadow probability model to obtain the shadow attributes while use threshold strategy to detect shadow. This algorithm is running on benchmark data set and gets a good approximation.The test data of this paper is from the benchmark data set and the transportation centers. In the data set, the background is complex; different illumination conditions, including shade, rain, and sunny days; the appearance of the pedestrian is different, low resolution, low contrast; in the scenes, a lot of nonhuman objects are moving, such as vehicles and dynamic backgrounds, etc. A large number of test results show that the algorithm feasibility of theory research.