Motion recognition method

The invention discloses a motion recognition method. The method comprises the following steps of: establishing a motion database; overlapping frame images in the motion database, and taking the overlapped frame images as input of an automatic encoder to train the automatic encoder; recognizing a neural network through a training mode so as to establish a deep neural network; and extracting a frame image of a to-be-recognized motion, comparing the to-be-recognized frame image with the deep neural network and outputting a recognition result. The method is capable of solving the problems that the time cost is high, the training time is long, the hardware requirement is high, the data storage space is big and timely capture is required in the prior art. 本发明公开了种运动识别方法，包括建立动作数据库；将动作数据库中的帧图像进行叠加，然后将其作为自动编码器的输入，对自动编码器进行训练；然后通过训练模式识别神经网络，建立深度神经网络；在通过提取待识别动作的帧图像，将待识别帧图像与深度神经网络进行对比，并输出识别结果。该方法解决了现有技术中存在的时间成本高、训练时间长、硬件要求高、数据存储空间大以及及时捕获性问题。