• A. Pazos
  • A. Rivas
  • R. Barral


INTRODUCTION - Since a few years ago computer science means an important sup port to biomechanical analysis. Whenever a lot of calculations are to be made, and the value of different parameters like mechanical variables are needed, the advantages of using computers are clear. However a bottleneck is present in data acquisition process for cinematic analysis from video sequence. Traditionally this task is performed like a manual process: user of computer systems must mark, for each frame, some articular points (about 20-21 ) by means of an optical pencil or using a mouse on the computer display once image has been digitalized. This is a routinary task and takes a lot of time. For example, for three seconds of movement, we probably need to process about 150 frames and, for each of them, to perform a digitalizing process. Recently new approaches are used to allow an automatic recognition. These approaches are based on the use of body optical sensors. Recognition process is easier because we only see several single points in the screen (under special environment conditions). However, these approaches are no applicable in real situations (i.e. competitions) which are the most interesting moments for analysis. What we propose here is an attempt to make an automatic system for data acquisition process from video sequences in sport environments and, as a general rule, for the analysis of human movement. We must take into account that the complexity of recognition process is lowered if we are working into a very narrow domain, like cyclic movements with 2D analysis (i.e. path of legs in some kinds of running). Nevertheless, our approach can be transferred to movements with 3D analysis if tridimensional reconstruction from human shapes has been performed. The process that our system will perform on each frame includes: . Digital image pre-processing, Edge sharpening, contrast adjust and filling of areas of interest. . If results are on satisfactoy (e.g. incomplete shapes), an Artificial Neural Network is used in order to predict total pattern, using previous frames or information available in a sportmen customized database containing antrophometric data and cinematic pattern of movement. . Last step is an intelligent matching between articular segments included in database (invariables botn with a 3D analysis and with a 2D analysis under specific features mentioned above) and human shapes obtained from previous steps. RESULTS - First phase relating to image recognition in laboratory conditions (high contrast) has been performed and the difference between manual articular coordenates and automatic acquisition was about zero-two units. CONCLUSION - The system we propose can help to biomechanics to reduce a lot the time destinated to perform data acquisition. Several hours would be changed to some minutes without human assist. REFERENCES R. J. Schalkoff, (1988) Digital lmage Processing and Computer Vision. John Wiley and Sons Inc. L.J. Galbiati, (1988) Machine Vision and Digital lmage Processing Fundamentals. Prentice-Hall International. C.H. Chen, (1993) Handbook of Pattern Recognition and Computer Vision. World Scientific.
Equipment / Instrumentation