Research On Dynamic Gesture Tracking And Recongnitin Method Based On TLD And HMM

Based on the research of dynamic gesture tracking and recognition,there are many uncertainties in data.Such as the rapid movement of the hand during the movement,the influence of the background,and the shooting angle,etc.In order to further improve the target tracking accuracy and dynamic gesture recognition accuracy,we need to research on dynamic gesture data,and then put forward a credible and effective method.The source of the dataset is collected from many people.Data features include fixed camera,different angles,deformation characteristics.According to the characteristics of video data flow for a long time,different angle.This paper puts forward corresponding tracking algorithm and recognition algorithm.And through the experiment,this paper completed the evaluation and analysis.The main contents and conclusion of this paper are as follows:(1)Dynamic gesture target tracking.According to the long and accurate tracking of dynamic gestures,this paper use TLD algorithm to track the dynamic gesture.According to the big error problem in the initial frame calibration,this paper use the candidate frame.The candidate frame selection method is obtained by calculating the spatial overlap.This paper selects the top 15 largest frame as the initial frame.According to the calculation of the error,this paper uses the normalized correlation coefficient and the sum of the squared differences.According to the problem of setting the threshold value in the inspection module,this paper uses the experimental method to verify selection.(2)Dynamic gesture target feature extraction.According to the problem with relative distances and deformations in dynamic gesture movements,this paper uses the gesture position point,movement speed and direction as the feature of dynamic gestures.The dynamic gesture position point is obtained by calculating the distance from the center point of the dynamic gesture track to any point in the track.The dynamic gesture movement rate is obtained by calculating the velocity of two adjacent center points.The dynamic gesture direction is obtained by calculating the angle between the track point and the center point,the angle between the initial track point and any track point,and the angle between the adjacent track points.(3)Dynamic gesture target recognition.According to containing a single data sample,this paper builds recognition training model based on hidden markov theory.The optimal parameters of the model obtained by training.Through the experimental analysis,the recognition accuracy rate is above 90%.According to the samples containing mixed data,this paper optimizes it on the basis of the existing model.This paper use weight parameters to training data in training model.This paper gets the best weight parameters.Through experimental analysis,the recognition accuracy rate increased to more than 91%.