Moving Object Detection In Motion Background Combined With Optical Flow And Saliency

With the rapid development of artificial intelligence technology,people's daily life has gradually entered the era of intelligence,and the society is increasingly demanding for intelligence.Moving object detection module plays a key role in intelligent monitoring system.However there are still some difficulties in the detection of moving targets,such as the complex movement of the background in the natural environment,and the inaccurate detection of moving targets caused by camera shake interference.In order to quickly capture moving objects from video sequences,this paper proposes a method of moving object detection combining optical flow method and salient features,which is mainly studied from the following three aspects:(1)According to the characteristics of dynamic background,the way of fixed optical flow field direction is studied to distinguish dynamic background and moving target.The traditional optical flow method can obtain the speed of an independent moving target without knowing any relevant information of the scene in advance,while the moving background generates a lot of background optical flow noise,which affects the detection accuracy.In this paper,the mean velocity in horizontal and vertical directions is used to represent the background light flow,and the motion characteristics of the object are obtained by calculating the difference.(2)Any moving target has its unique characteristics.Based on this characteristic,this paper proposes a saliency detection algorithm for moving targets in a dynamic background.The bottom-up saliency detection algorithm ignores the spatial correlation between the underlying features,resulting in obvious noise and blurred features.Therefore,this paper proposes a saliency detection method based on reconstruction and segmentation to extract saliency feature information to ensure the completeness of the target area information: Firstly,the structural elements in image reconstruction are used to suppress the high-frequency noise information in background information;Secondly,the pixels with similar characteristics are clustered and segmented according to the correlation of color spatial distribution between pixels;Finally,the significant value is calculated.(3)Due to the complexity of the dynamic background,a single optical flow field or saliency detection method for a moving target is difficult to meet the requirements of stable detection.In order to improve the detection accuracy,this paper combines the optical flow field and saliency to carry out research on the detection method of moving targets under dynamic background: First normalize the optical flow result image and the salient result image,then perform matrix point multiplication on the two feature images,and finally perform dynamic weighted fusion.The experimental results show that the detection effect of the single method of dynamic background is easily interfered by complex moving backgrounds,and the optical flow feature vector combined with salient features can well extract the moving target information from the complex moving background,which can completely retain the target area,reduce the impact of background motion,and has good stability and robustness.