The cow body size is an important indicator for evaluating the health status of the cow, and the extraction of body measurement point is critical in the body measurement process. In order to solve the problem of automatic extraction of the measuring point of the cows point cloud, a method for automatic extraction of the measuring point of the back of the cow was proposed. Firstly, a cow deep video capture platform for data acquisition was built, the original point cloud data of the cows back collected by the Kinect camera was preprocessed to remove the surrounding complex background. Secondly, the principal component analysis method was used to calculate the local plane normal vector and curvature, the point cloud of the cow back was streamlined, the noise points and redundant points were removed, and the characteristic points of the cows back ridge and boundary contour were preserved. Finally, according to the geometric characteristics of the measuring point of the back of the cow and the spatial structure relationship between the measuring points, the body point measurement data of the cow’s back point cloud data were automatically extracted. Based on the complete back depth video of 33 cows, ten frames of each cow were selected for a total of 330 frames of experimental data, and the average absolute error of all body measurement points extracted by this method was less than 1.17cm, which can meet the requirements of measurement error of the back of the cow in the livestock application. Compared with the traditional streamlined method, the extraction time of the body measuring point after the simplified method was shortened by 33.72%. The results of this study can provide technical support for automatic measurement of cow body size.