人体运动轨迹, 图像分割, 七帧对称差分算法, Mean Shift 算法

In the actual scene, there are similar colors and textures between the human body and the background, and the movement of the human body involves diversity of gestures. In this complex and changeable background, it is difficult to segment the trajectory of the human body. Therefore, an image segmentation technique based on symmetric difference algorithm is proposed. The seven-frame symmetric difference algorithm is used to extract the first three frames and the last three frames of the human motion image sequence, the absolute difference images are calculated, and the human motion target region is obtained. A non-parametric statistical iteration (Mean Shift) algorithm is used to extract the distribution of pixel modulus points and generate superpixels. A non-parametric Bayesian clustering model is used to fuse superpixels and to extract the contours of human moving objects. Gaussian mixture model is used to establish human trajectory model, and extreme learning machine is used to solve the model recognizing human trajectory and realize human trajectory image segmentation. The experimental results show that the IOU ( Intersection Over Union) value of the proposed method can reach up to 97% , and has high precision of extracting moving target region, high precision of identifying moving trajectory and good segmentation performance, and is suitable for human motion trajectory image segmentation.

human motion trajectory, image segmentation, seven frame symmetric difference algorithm, Mean Shift algorithm