Research Method of Blind Path Recognition Based on DCGAN

In order to solve the problem that there are few blind path data sets and a lot of manual data collection work in the current blind guide system, computer vision algorithm is used to automatically generate blind path images in different environments. Methods a blind path image generation method based on the depth convolution generative adversary network (DCGAN) is proposed. The method uses the characteristics of typical blind path, which is the combination of depression and bulge. The aim of long short memory network’ (LSTM) is to encode the depression part, and the aim of convolution neural network (CNN) is to encode the bulge part. The two aspects of information are combined to generate blind path images in different environments. It can effectively improve the blind path recognition rate of the instrument and improve the safe travel of the visually impaired. Conclusion generative adversarial networks (GANs) can be used to generate realistic blind image, which has certain application value in expanding blind channel recognition data, but it still needs to be improved in some details.