Abstract: The recognition rate of gesture recognition is low because of the poor segmentation effect. Therefore, a dynamic multi-point gesture recognition method based on improved support vector machine is proposed. The depth threshold method is used to segment the dynamic multi-point gesture image, extract the largest circular fine hand area in the palm, obtain 7-dimensional HOG(Histogram of Oriented Gradients) feature vector of the hand, complete the gesture action image preprocessing, introduce support vector machine, and improve the algorithm by error term, and adopt the optimized linear classification feature vector of the improved support vector machine. The dynamic multi-point gesture recognition is realized by using the gesture feature vector after input classification by support vector machine. The experimental results show that the recognition rate reaches more than 92. 5% under the condition of illumination, while the recognition rate is still higher than 90. 0% under the condition of no illumination. The proposed method has little fluctuation under the influence of illumination, and the image segmentation information is complete and the recognition accuracy is high.

Key words: improved support vector machine, dynamic multi-point gesture, gesture recognition, histogram of oriented gradient(HOG) feature extraction, back propagation(BP) neural network