Research On The Technology Of Dynamic Gesture Recognition Based On The Integration Of DTW Algorithm And Optical Flow Method

With the popularity and development of the proposed human-computerinteraction techniques (HCI),the technology in dynamic gesture recognition isbecoming significant. Studying dynamic target tracking in the complex background isnot only of great theoretical significance, but also has a wide and high practical value.The recognition system of Dynamic gesture in the paper mainly consists of fourparts，the image preprocessing and segmentation, the learning and extraction of handfeature, the trajectory feature extraction, the training and recognition based onDTW(Dynamic Time Warping) algorithm. In the part of gesture segmentation, italways used the segmentation based on skin color. However, when the background isrelatively complicated, there is no way to track the position of the hand correctly. Thecentroid estimation can solve the problem that effectively track the movement processof hand. In the part of dynamic gesture trajectory tracking process, the sparse opticalflow, instead of the centroid estimation, can solve the problem that leads an unstablegesture trajectory tracking by the complexity of the background. The method canprocess the image directly without pre-acquiring the background, and the result onlyrelays on the relative motion of successive frames without the effect of complicatedbackground. In the part of dynamic gesture recognition, it uses DTW for its easytraining, while it brings the Computational complexity. Here, it is proposed themethod that controls the matching scope and degree of distortion of two time-seriesdata to optimize the algorithm.Its innovative is mainly shown in the following three aspects:(1)This paper using the motion features of gesture, introduced a centroidestimation in gesture segmentation methods, because the existing methods can’tsegment the image in the complexed background. In this method, Hand and faceregions are extracted based on skin color segmentation. Then it uses both the handcentroid and the motion features of other regional (facial) features centroid to split thehand region out. This segmentation is easy and efficient, with better experimentalresults.(2)It is proposed sparse optical flow method based on iterative LK (Lucas Kanade)pyramid for effective trajectory tracking in this paper. This method through theanalysis of the movement of the hand’s key position to track gestures, solves theproblem that the centroid is vulnerable to the region with similar skin color in the original centroid tracking method. Now, the improved sparse optical flow methoddoes not only reduce the analysis of the image content, but also improve the trackingaccuracy.(3)The paper proposed the optimized DTW algorithm to effectively control thematching scope and the distortion of the two time-series data. It reduces both thecomputational complexity and storage space, and improves the efficiency of thealgorithm. Thereby, it enhanced the real-time performance of the algorithm.Experimental results show that the method adopting the fusion of the sparseoptical flow method and the optimized DTW matching algorithm, promoted theaccuracy and speed of dynamic gesture recognition.