Application Research Of RPCA Model Combined With Foreground Spatio-Temporal Information In Moving Object Detection

Moving object detection is an important research content in the field of computer vision and artificial intelligence.Because the real scene is affected by external factors such as climate and illumination,the background environment will change to different degrees,and the object detection algorithm based on the background model can not achieve the ideal detection effect.Robust principal component analysis(RPCA)based on low rank and sparse decomposition representation has become the mainstream method for moving target detection,However,when the RPCA model is applied to the detection of moving objects in the dynamic background,there are some problems such as weak anti-noise and incomplete extraction of objects when the object and background pixels are similar.In this thesis,the RPCA model is deeply studied,and the motion consistency detection method is combined with the IALM algorithm for solving the RPCA model to improve the anti-noise ability of the model;the foreground connectivity and boundary connectivity metrics are combined with the initial prospect estimation of the foreground connectivity and the foreground weight calculation method are used to improve the integrity of the target extraction.The main work of this paper is as follows:(1)Research on the algorithm of solving RPCA model,focusing on IT algorithm,APG algorithm,EALM algorithm and IALM algorithm.Then the algorithm which is feasible for the latter three is tested and compared in static background.The experimental results show that all three algorithms can accurately detect the target contour,and the IALM algorithm has the highest efficiency.Then the IALM algorithm is used to detect the target in the dynamic background.It is found that the anti-noise ability of the IALM algorithm is not strong.In response to this problem,combined with the time information of the foreground,the motion consistency detection method is introduced and improved,and the dynamic background noise is solved and improve the accuracy of RPCA model for dynamic background target detection.(2)Aiming at the problem of incomplete extraction of foreground target in RPCA model,combining the spatial-temporal information of the foreground,the result obtained by the IALM algorithm and the motion consistency test is used as the initial prospect of the foreground connectivity metric,the foreground connectivity is calculated and the foreground weight calculation method is improved.The calculation method of the weight is improved.Then the background weight is obtained by using the boundary connectivity measure,and then the foreground weights are combined with the background weights to obtain a more complete object detection result.(3)Using the CDW2014 and UCF-Sports datasets,the effectiveness of the object detection algorithm studied in this thesis is validated,and the recall rate,accuracy rate and F1 score are selected to evaluate the effectiveness of the algorithm.The method in this paper can extract foreground targets completely under dynamic background and strong illumination.