Based on the CNN model, a gesture recognition algorithm named Add_Layer_CNN (A_L_CNN for short) is proposed,

which is a three-layer extraction and anti-overfitting gesture recognition algorithm based on neural network CNN. A_L_CNN changes the single-layer convolution pooling in the traditional CNN model to three-layer convolution cubic pooling structurally, and adds the Dropout (random deactivation) layer to prevent overfitting. A_L_CNN is compared with traditional CNN and SVM. Experimental results in multiple test sets show that the average accuracy of the proposed A_L_CNN model is about 98.56%, that of the traditional CNN model is about 96.11%, and that of the SVM model is about 87.25%. Therefore, the accuracy of the proposed A_L_CNN model is higher.