In order to efficiently and automatically generate the Laban dance spectrum for the performer's actions in the non-heritage performance video, and reduce the learning cost of the dance spectrum plotter, the action extraction and Laban dance spectrum generation system in the non-heritage performance video is designed and developed. This system is established in the background of ordinary non-legacy performance video, mainly for the recorded two-dimensional video dance spectrum automatic generation, which does not contain audio information, depth information, etc., at the same time, there is no camera internal reference, height, angle and other information, and no additional calibration reference point and other operations. First, train the system to study the improved human body two-dimensional key point detection and three-dimensional attitude estimation algorithm; then, input the non-heritage performance video to be recorded into the trained algorithm model; then, detect the two-dimensional key points of the performer's limbs, and then input the results of the two-dimensional key points of the performer into the three-dimensional attitude estimation algorithm to realize the three-dimensional attitude estimation of the non-heritage video performer and visualize it on the result display interface; finally, the predicted three-dimensional pose result of the performer is input into the Laban dance score generation algorithm, and the Laban dance score is automatically generated and displayed in the result display page.