Abstract:The development of the depth camera makes it more convenient to achieve gesture skeletal information. To obtain useful information from the big data of multidimensional gesture skeletal nodes and realize the recognition of common twohanded staticinteractive actions in the complex indoor environment and close range conditions, a static gesturerecognitionmethodis proposed.Itisbasedon multifeaturefusionand multiclassification supportvector classifier(multiclassSVC). To achieve better results,multiclassSVCis optimized by the bioheuristic genetic algorithm. By using gesture skeletal data, a new gesture feature is designed and a more comprehensive gesture feature sequence is established through the feature combination strategy. In this way, the influence of redundant features is reduced and the ability of data processing is enhanced. The optimal kernel function and penalty parameters are obtained by optimizing the kernel function and penalty parameters of multiclassSVC with the bioheuristic genetic algorithm. The issue of low gesture recognition accuracy is addressed, which is caused by the random selection of the kernel function and penalty parameters. Comprehensive evaluations of the gesture recognition model are carried out by using P, R, F1 and A. Comparison experiments with KNN, MLP, MLR, XGboost and other models verify that the proposed gesture recognition model can effectively improve the accuracy of gesture recognition. This paper analyzes the influence of sample size on gesture recognition accuracy through model training by adding gesture sample data iteratively. It provides an effective method to improve gesture recognition accuracy. Experimental results show that the gesture recognition accuracy can reach 984%. And the average precision rate, recall rate and F1 performance evaluation indexes of the recognition algorithm are not lower than 098.