The Key Pose Detectionand Motion Analysis In The Weightligting Videos

The explosion extansion of sport videos proposes the higher requirement tovideo analysis. Introduction of the intelligent computation technologies to sport videoanalysis could improve the performance of video analysis. It not only bringsconvenience for the audience to watch the sports programs, but also provides aneffective tool for coaches, and athletes,and make the training more effective.The thesis focuses on the motion analysis of weightlifting videos. According therequirement from the training, the video is analyzed statistically. And the key pose isextracted automatically. The details are as follows:1、We propose a foreground object detection method combining the inter framedifference with logical "or" operation. It generates a reliable detection range for thesubsequent key pose detection and optical flow statistical analysis. This processreduces the background interference and improves the speed.2、The pyramid LK optical flow method is applied to weightlifting video. Fourchannel descriptors are established to analyze the statistical properties of trackingresults. We found the connection between the extreme point of the descriptor channeland the key pose frames. So the statistical property of the motion is applied to detectthe key pose frames.3、A poselet based scheme is proposed to detect the key poses in the weightliftingvideo. First, the poselet based SVM classifiers are obtained by two training stepsusing the poselet samples and a large number of negative samples. The frame containsthe most detection hits is selected as the best key pose frame under the multi-scalescanning mode. This rule solves the problem of inter-frame similarity effectively。...