Research On Moving Target Detection Method In Complex Scene Based On Background Modeling

Intelligent video analysis technology combines machine vision and traditional monitoring systems,and it is an important research direction in the field of artificial intelligence and machine vision,which is widely used in national public security,intelligent transportation network,intelligent power,building intelligence and other fields.Intelligent video analysis technology needs to establish the mapping relationship between the video and its description,and understand the semantic information of the original data through technologies such as machine vision,pattern recognition,image processing and artificial intelligence.Moving target detection plays an extremely role in improving the calculation speed,accuracy and robustness of the intelligent video analysis system,which directly affects the overall performance of the system.The background modeling method is the most commonly used method in the field of moving target detection.A stable background model is established through video sequences,and the pixel value,histogram,texture and other characteristic information in the established background model and the corresponding features of the input image are obtained by differential operation Sports goals.The core of the background modeling method is the establishment and update of the background model,which has the advantages of simple principle,fast calculation speed,small amount of calculation,high real-time performance and high target detection accuracy.In this thesis using machine vision and other related theories,the recognition and optimization of motion targets in complex video scenes and sudden changes in lighting scenes are thoroughly studied,and proposes the following four background modeling methods.(1)In order to improve the accuracy of moving target detection and construct a robust background model,a background modeling method based on multiple feature fusion is proposed,which based on the comprehensive consideration of the temporal correlation of pixels in the same position of the video image and the spatial correlation of neighboring pixels.In the background modeling method,four features in the video sequence are selected to establish a background model.Firstly,the first frame of the video is collected,and the background model is quickly initialized by using the correlation characteristics of pixels and their neighboring pixels.Secondly,the background model is updated by fusing multiple features of the video sequence to quickly eliminate the ghost area generated by the background model initialization when detecting the target.Finally,by extracting the pixel value of the input frame and the feature value in the background model,the adaptive threshold is adjusted to improve the adaptability and detection rate of the background model in the detection of targets in complex scenes.(2)In order to solve the problem that the background modeling method is very difficult to detect the target when the pixel value of the whole frame changes widely under the sudden change of illumination,we propose a multi-feature background modeling method based on the improved Census transform.Firstly,the median value of the Census transform window is selected to replace the center pixel value to reduce the dependence of the Census transform feature value on the center pixel.Secondly,the Census template and update template rules are established,which effectively improves the Census The transform is robust and stable to video processing.Finally,the improved Census transform and other features are merged to establish a background model to realize moving target detection,improve the accuracy of target detection in scenes of sudden light changes,and use adaptive sensitivity coefficients Measure the dynamic degree of different areas of the video and set corresponding update rules,which improves the robustness of detecting moving targets in complex scenes.(3)Most background modeling methods are sensitive to the complex changes of the background,especially in complex scenes,where large dynamic changes in the background often cause errors in detecting foreground targets and recognizing the background,resulting in a low detection rate.A background modeling method based on adaptive complex scenes is proposed.The background model is initialized by collecting the first 5 frames of the video,and the background model is updated by the pixel information obtained from the input frame,which reduces the influence of noise and intra-frame edges on the detection target,and improves the initialization that is easy to cause ghosts.The problem of shadow phenomenon.In view of the large dynamic changes in complex scenes,it is difficult to directly describe the features with pixel values,and the adaptive discrete coefficients are used to describe the dynamic changes of pixels,which reduces the difficulty of extracting foreground targets in complex scenes and reduces false foregrounds caused by large changes in background pixel values.(4)For most background modeling methods,the detection of moving targets is prone to false targets,foreground holes and ghosts.A background modeling method based on improved morphology is proposed.Firstly,moving targets are quickly detected based on the improved Vi Be algorithm.Secondly,the background pixels are divided into two types of simple background and complex background regions by dynamic coefficients to constrain the corrosion operation range,which reduces false foregrounds.For the impact on the detection result,the expansion structure element is improved and the target hole is filled.By limiting the area of corrosion and expansion,the frequency of execution of corrosion and expansion in the video sequence is reduced,and the influence of morphology on the speed of model detection is reduced.Finally,a method for detecting and eliminating ghost regions based on the median of the neighborhood is proposed,which can suppress the ghost regions well.