Research On Algorithms Of Human Detection In Videos

In the field of computer vision area, pedestrian detection has a large applicationprospect. The person’s appearance, actions, and events information will be transmittedand processed via the advanced man-machine interactive interface and then theinformation will be use to create realistic human body models. Finally these detectiontechnologies can be applied to various fields. Such as: smart car, video surveillance,image retrieval and advanced human-computer interaction. Obviously, the methods ofimage pattern recognition can’t meet the high-level application requirements. This articleis trying to meet the purpose of pedestrian detection in videos.In this article, pedestrian detection process consists of two sub-processes: movingobjects detection and human detection. In the moving objects detection section,Background subtraction was used for detecting moving objects. In order to solve theproblem of background changes over time, we put forward Gaussian mixture model andupdate the background data in real time. In order to learn the non-rigid deformation ofthe body feature, we selected Haar-like features and AdaBoost training method to get thebody detector.Besides, we modified the foreground image with mathematical morphologyprocessing after convert the foreground image to binary one, the8-connected componentlabeling algorithm was used in this step. This method greatly reduces the pedestriandetection area, improves the accuracy of detection and ensures the detection speed.Experimental results shows that the method proposed can be used for real-timehuman detection.