Characteristic point recognition method based on neural network

The invention provides a characteristic point recognition method based on a neural network. The characteristic point recognition method comprises the following steps: training the neural network by utilizing a pre-processed picture; keeping an infrared spot light source of a virtual reality helmet in an open state, and photographing by an infrared camera; pre-processing an image to obtain the pre-processed image; inputting the obtained pre-processed image into a neuron to obtain an ID (Identity) of the infrared spot light source corresponding to each light spot. Compared with the prior art, an algorithm of the neural network is introduced into a virtual reality space positioning method, and a method for determining the ID of the light spot is provided and is accurate and efficient. A training image and a testing image are pre-processed, so as to avoid influences, caused by the diversification of pictures, on recognition accuracy; the diversified pictures are subjected to standardization processing, so that the succuess rate and accuracy of ID recognition are greatly improved.