基于GMM的人体运动姿态的追踪与识别

Human Motion Tracking and Recognition Based on GMM

随着人工智能等技术的兴起,利用机器视觉对视频中运动目标进行追踪与识别在工业、交通、医疗和运动训练等领域都得到应用.对视频中人体运动姿态进行准确快速的检测,是目前一个热门的研究方向.本文采用改进的混合高斯背景模型(GMM)算法对视频每帧图像进行前景提取,通过帧间差分法分析得出不同差值对应的学习率,从而实现对背景模型更准确的更新,进而得到一个精确的二值化的前景图像;并将生成二值图像由更新后的像素与高斯B均值比较,得到背景或前景图像;再对处理后视频图像进行比对,利用Shi-tomasi算法提取图像特征点并进行追踪,获取运动目标轮廓并绘制出边缘,经过SVM训练实现对走、跳、跑3种人体运动姿态的实时追踪和识别.

With the emergence of technologies like artificial intelligence,tracking and recognition of moving objectives in the video with machine vision have been applied in the fields such as industry,transportation,health care and athletic training,etc.. Currently,accurate and rapid detection of human posture during motion is a hot research interest. In this paper,the foreground was extracted from each frame of the image using an improved algorithm for a Gaussian Mixture Model（GMM）,and the learning rates corresponding to different differences were obtained through an inter frame difference method of background models were upgraded more accurately and a precise binary foreground image was acquired; a background image or a foreground image was generated from binary image after the comparison of upgraded pixels and the average value Gaussian B; the processed video images were compared and image feature points were extracted using the Shi-tomasi algorithm and traced to get the contour of moving objectives and the edge was plotted,and three human motion postures walk,jump and run were real-time tracked and recognized by SVM training.