Research On Video Pedestrian Detection Based On Background Model And Fast Feature Pyramid

The pedestrian detection is a research hotspot in the computer vision,and it has a wide range of applications in video surveillance systems,unmanned vehicle and intelligent transportation etc.For video pedestrian detection,the motion detection combined with the pedestrian detection is usually used in practical applications.Firstly,the motion detection is performed on the video sequence for extracting the motion area,and then the pedestrian detection is performed on the extracted motion area.Compared with the traditional pedestrian detection in each frame of the video sequence,the method of motion detection combined with pedestrian detection has higher detection accuracy and detection speed,and it is more suitable for realistic scene.In this paper,the motion detection and pedestrian detection are studied separately.The main research work is as follows:Firstly,based on the PAWCS algorithm,this paper introduces the shortcomings of the LBSP feature that the foreground detection sensitivity is too high and the detection results on the intermittent motion target is still unsatisfactory.The update method of the background model is improved.The quadratic neighborhood update and the suppression of foreground motion pixels update are added.A PAWCS+ algorithm with improved model update mechanism is obtained.The detection results of PAWCS+ and PAWCS are analyzed through experiments.The experiments show that the PAWCS+ algorithm proposed in this paper solves the problem of PAWCS algorithm to some extent and improves the detection results of PAWCS algorithm.Secondly,based on the ACF algorithm,this paper introduces the shortcomings of the single average downsampling for the gradient amplitude channel feature.The downsampling method for the gradient amplitude channel feature is improved.A quadratic downsampling method based on stochastic downsampling and average downsampling is used to obtain a BACF algorithm.The detection results of BACF and ACF are analyzed through experiments.The experiments show that the proposed BACF algorithm improves the detection results of the ACF algorithm to some extent.Finally,based on PAWCS+ algorithm and BACF algorithm,a video pedestrian detection algorithm is proposed.The algorithm is used to optimize by the motion target re-expansion and noise motion region pre-filtering.The detection accuracy and detection speed in practical applications is improved further.In this paper,the detection results of the video pedestrian detection algorithm and the BACF algorithm are analyzed through experiments.The experiments show that the proposed video pedestrian detection algorithm has a better performance than the single BACF algorithm.