Research On Key Technologies Of Semantic Calculation Of Sports Human Behavior

The human motion's behavior recognition based vision has extensive application prospect,but most of these achievements are about the template matching method and State Space Method which have Semantic Gap with human Comprehension,the research achievements about semantic way which can overcome this fault is very few.The semantic compute method which fuses scence,human characters and motion is studied in this paper.For achiving this goal, the paper main works as follows:(1) A novel improvement particle filter algorithm for 3D human tracking is proposed in this paper which based Parallel Computing in Beowulf cluster system. The new algorithm can realize automatic recovery from tracking failure by automatically initializing 3D human body model Parameters and adjusting of particle amount and template. The new way also can solve particle degeneracy problem and improve computation speed by migration particle filter parallel algorithm which based task dynamic allocation and low consumption communication strategy in Beowulf cluster system.(2) For improving scene recognition rate and extracting high level semantic, a new way of scene semantic recognition based four layers treelike semantic model is purposed. The four layers semantic model includes visual layer and the concept layer and relation layer and semantic layer. The visual layer is obtained by extracting the color and color gradation and outline of scene entity in training sample. The concept layer is constituted by intersection of the same kind scene entity name which is concept words. The relation layer is obtained by counting the frequency of concept words and data mining of concept words space location relationship association rules. The scene high level semantic is gotten by calculating the semantic similarity between PSB standard semantic attribute classification trees and key concept words. After calculating the low-level image features, the concept words are gotten by searching visual layer. The scene classification is obtained by searching the frequency and space location relationship association rules of concept words. The scene semantic recognition result is constituted by scene classification and scene high level semantic and concept words.(3)For improving identification rate and reducing identification time, a cognitive physics method of human identification is presented in this paper. The facial features and gait features are described by data field, the interaction and motion among datum is used to realize self-organization cluster of datum which is nonlinear conversion way for reducing the dimensions of identification features datum. The dimension reduction of sample database is sorted by maximum potential value so that the rapid detection of discrete point and Two Searching Method of samples testing are realized. The facial features and gait features are fused based improvement D-S evidence theory.(4)The existing semantic method has not integrated scene semantic and human identity and lacks the effective recognition and description for complex behavior, a new semantic recognition way of human behavior in video images which based hierarchical concept space is proposed.The new method introduces concept space in cognition science and has built hierarchical motion concept space, the complex behaviors are decomposed as atomic motion layer and simple behavior layer and event behavior layer. The atomic motion layer is subdivide body movement and limb and posture layer atomic motion then human moving features are extracted by large scale and mesoscale and small scale. The atomic motion detection is realized by concept activation function and simple semantic recognition is realized by space-time logic rules. The seven tuple semantic model is proposed which can merge into scene semantic and person identity and realize event behavior semantic modeling and event behavior semantic recognition.