Research And Development Of Auxiliary Navigation System For The Visually Impaired Based On Semantic Segmentation

China is the country with the largest number of blind people in the world.As a social vulnerable group,blind people need to attract the attention and care of all sectors of society.In the daily life of the blind group,due to the limitation of visual function,the most difficult challenge is to travel safely.With the improvement of social security system,some auxiliary travel tools have appeared in the market to solve the travel needs of the blind.However,when we need to deal with the travel problems in the complex environment such as large flow of people,imperfect traffic guidance facilities,and sidewalk obstruction,most of them need to wear additional or hand-held devices which can not achieve the functions of convenient use,obstacle danger identification,warning and other travel safety protection at the same time.In this paper,the related products and technologies are investigated in detail.Using the development advantages of deep learning in the field of computer vision,the auxiliary navigation system for the visually impaired based on semantic segmentation is developed.The main work of this paper is as follows:(1)In order to solve the problem that the target environment of urban street scene is complex and the analytical ability of traditional multi-scale detection algorithm is limited,a new semantic segmentation model,multi-scale deformable convolutional network is proposed.By using deformation convolution,that is,adding an offset,the model can adapt to complex and irregular geometry.Then the input image information is processed on multiple scales and different levels of detail in the scene are captured.Experimental results show that the Miou accuracy of the new network model is 80.2%on cityscapes dataset.(2)Based on the analysis of the existing travel mode of the blind group,we design an auxiliary navigation scheme for the visually impaired based on semantic segmentation.The solution relies on Android system and uses react native cross platform development framework.The application integrates voice recognition and environmental obstacle detection.We designed and developed voice interaction module,obstacle monitoring module and route navigation module.Finally,a low-cost and convenient travel plan is achieved.After testing and verification,the system realizes obstacle detection,road navigation and other auxiliary navigation functions,and achieves the expected goal.