Research On Moving Object Detection Based On Video Sequences Of Complex Environment

Moving object detection is the key technology in the field of machine vision.Our paper mainly studies that background modelling in the different complex environment,moving object detection in stationary complex environment,and moving object detection in dynamic complex environment.In order to deal with the influence of shadow interference,texture deficiency and scene jitter on the background in the stationary complex environment,the corresponding background image representation models are established.Aiming at the influence of rotation,scale and viewpoint changes on the background under the dynamic complex environment,the background transformation models corresponding to different changes are studied.To eliminate different disturbances in the stationary complex environment,a background modeling and object detection algorithm is proposed which is based on local fusion feature and Gaussian mixture model.The algorithm proposes improved local fusion feature and uses Gaussian mixture model as base.It exploits multiple decision to learn and update the background model parameters,which improves the robustness of itself.In order to make model parameters adapt to the complex background,a background modeling and object detection algorithm based on variational Bayesian learning is proposed.The above model is based as the basic Bayesian framework in this algorithm.Its model distributions are optimized by online expectation maximization algorithm to automatically estimate the optimal number and parameters of the Gaussian model.Experiments show: the improved detection algorithm based on local fusion features can effectively deal with the different interference in the stationary complex environment.The algorithm has strong robustness and high precision.Combined with variational Bayesian learning,the object detection algorithm can keep strong robustness and improve the accuary of object detection further.To resist different changes in the dynamic complex environment,an object detection algorithm based on joint matching tracking and regional dual-mode Gaussian modeling is proposed.The object detection accuracy is improved by background compensation algorithm of joint matching tracking and background modeling of regionalized dual-mode Gaussian function.ORB algorithm is improved to obtain the matching points;LKR(Lucas Kanade Random)algorithm is proposed to obtain the tracking points;the joint matching points and the tracking points are used to calculate the confidence feature points;the background model is regionalized and modeled by dual-mode Gaussian function which is combined by online model and candidate model.Experiments show: the proposed algorithm achieves good object detection performance in the video sequence of dynamic complex environment under the above background changes.It has high robustness and accuracy.The average detection frame rate is 40.6 fps,which meets the real-time requirement.