Faster R-CNN-based human pose estimation method

The invention discloses a Faster R-CNN-based human pose estimation method. The method comprises the following steps of: inputting an image; classifying human parts; obtaining human pose image data andlabels; training a deep network Faster R-CNN model by using training set image data and labels; obtaining a rectangular detection box; determining human part positions in a space constraint relationship; determining joint point positions; and connecting and outputting joint points of adjacent human parts so as to obtain a pose of an upper body. According to the method, the human parts are dividedinto single parts and combined parts, Faster R-CNN is adopted, and position coordinates corresponding to necks are taken as standards, so that high-precision upper body pose estimation can be obtained under the interference of image backgrounds. The method has the advantages of carrying out robust, high-precision, wide application scene human pose estimation. 本发明公开种基于Faster-RCNN的人体姿态估计方法，其步骤为：输入图像；进行人体部件分类；获取人体姿态图像数据与标签；用