INTRODUCTION: To test and analyze the kinematics of elite athletes in their natural training or competition environment represents one of the most interesting and biggest challenges of sport science nowadays. The state of the art for motion analysis in this field is marker tracking based on a video image or infrared (stereophotogrammetry). Besides the known problems of tissue artifact, displacement of markers, and alteration of motion patterns caused by markers, the laboratory measurement environment required for this method contradicts the demand of biomechanical analysis close to training or competition. Therefore, markerless video-based tracking is seen as a potential method to make movement analysis quicker, simpler, easier to conduct, and closer to the athlete’s familiar surroundings. Since this technology is still new for application in the fields of sports, markerless systems have to be evaluated for adequate accuracy. So far, There is only few scientific work dealing with the development and realization of silhouette-based tracking in gait analysis, and no general evaluation of silhouette-based tracking which takes all relevant kinematic parameters into account axist. The purpose of this study was therefore to assess the accuracy of the new silhouette-based tracking software Simi Shape compared to the 'gold standard' marker-based tracking. METHOD: In order to quantify the accuracy of silhouette and hybrid tracking against marker-based tracking, four different power strokes (serve, forehand, one-/two-handed backhand) of an elite tennis player were recorded by an 8-camera motion capture system (Simi Reality Motion Systems, Unterschleißheim, Germany). One and the same recording of each stroke was used for all three tracking methods. The marker-based tracking was based on a full body marker set with 46 retro reflective markers. 3D data of 12 markers attached to the upper limb and the trunk were used for the hybrid tracking. All in all, 1640 data pairs based on the record of the four tennis strokes were entered into the statistical analysis. The empirical correlation coefficient r and the standard deviation of the differences SD_Diff were compared for the parameters joint angle (JAng), joint center location (JC) and segment center of mass loacation (SegCOG). Furthermore, the gradient a_1 of a linear function as a result of a simple linear regression was used to check the similarity of data in case of a very strong correlations. RESULTS: The SegCOG as well as the JC reached a high level of tracking accuracy for each segment / joint in each stroke for silhouette tracking as well as for hybrid tracking. The overall standard deviation of the differences was SD_Diff = 27mm for the SegCOG and SD_Diff = 29mm for the JC. Hybrid tracking improved the SD_Diff for all SegCOG (SD_Diff = 19mm) and JC (SD_Diff = 17mm) of the upper trunk and upper limb. Joint angles showed very strong correlations for the ankle, knee and shoulder joint in the sagittal plane for the groundstrokes as well as for the shoulder abduction/adduction for all strokes. The SD varied for each joint of the upper body between the strokes and could not kept continuously below the defined threshold of 10°. The SD of the ankle joint and knee joint in sagittal plane for all strokes was smaller than 10°. Only the knee joint flexion/extension as well as shoulder abduction/adduction met all requirements of high tracking accuracy. For the elbow joint flexion/extension as well as for the left shoulder flexion/extension high accuracy results of hybrid tracking were gained for all stroke types. Furthermore, the correlation coefficient was improved for all shoulder joint movements. DISCUSSION/CONCLUSION: Due to the known errors of marker-based tracking, it's use as reference is questionable. Thus, small deviations between the data can either be based on tracking problems of markerless tracking or marker-based tracking. Furthermore, a clear separation between the errors based on the tracking method and those based on the different body models cannot be guaranteed. From an applied point of view, the presented results support the usage of this time saving tracking technology. Based on the good accuracy results for the JC as well as for the SegCOG, all scientific problems dealing with the body posture and movements of several resp. all segments and, thus, dealing with timing of single motion patterns and momentum transfer can be answered unrestrictedly. However, for exact joint angle data except the knee and elbow joint, additional sources of rotational information should be used. Due to the problems of marker-based tracking and the difficulties in separating effects based on the tracking method and effects based on the body mode, an alternative method should be used for the exact evaluation of tracking accuracy.     «

INTRODUCTION: To test and analyze the kinematics of elite athletes in their natural training or competition environment represents one of the most interesting and biggest challenges of sport science nowadays. The state of the art for motion analysis in this field is marker tracking based on a video image or infrared (stereophotogrammetry). Besides the known problems of tissue artifact, displacement of markers, and alteration of motion patterns caused by markers, the laboratory measurement enviro...     »