Research On Tactile Pavement Detection Algorithm Based On Deep Learning

Tactile pavement is one of the most important facilities to assist visually impaired people to travel safely,and tactile pavement can provide guidance and cues for visually impaired people.In reality,tactile pavement is complex,and visually impaired people have difficulties in identifying tactile pavement.Therefore,how to assist visually impaired people inaccurately identifying tactile pavement is an important part of research to ensure the safe travel of visually impaired people.Within the field of computer vision,image segmentation is one of the important elements to perform target detection and recognition.With the rapid development of computers,it has become possible to segment tactile road surfaces using image processing-related techniques to assist visually impaired people to travel safely.In this paper,we address the problems that visually impaired people have difficulties in traveling in real life,the complex tactile pavement situation outdoors,and the acquired tactile pavement images are prone to shadows and noise,resulting in low accuracy of tactile pavement segmentation.In this paper,we propose a tactile pavement segmentation method based on the color characteristics of tactile pavement in outdoor situations,converting the original image into YUV color space,extracting the shadow region in the image,and performing chromaticity correction based on the Y component representing the luminance information.The OTSU algorithm is used to segment the image after the de-shadowing process to initially extract the tactile pavement region,and the image is labeled by combining the distance transform,and the markers are used for tactile pavement segmentation by the watershed algorithm.The experiment proves that the method can avoid the influence of the shadow part in the image on the segmentation algorithm and improve the accurate efficiency of the tactile pavement segmentation.Tactile pavements in real life come in a variety of colors and are easily influenced by the surrounding environment.Tactile pavements have distinctive color and texture characteristics.Visually impaired people mainly find tactile pavement based on the tactile sensation of their feet,and can segment the tactile pavement based on its texture features.Firstly,the texture features of the image are extracted by Gabor filter,then the image is segmented by combining the K-means clustering method,and finally the segmentation of tactile pavement is completed with morphological processing and finding the maximum connected domain.The experiment proves that the method can segment tactile pavement images with complex backgrounds and contribute to the safe travel of visually impaired people.To address the problems of low accuracy,slow speed,and poor generalization of traditional tactile pavement segmentation algorithms,this paper proposes an improved Seg Net model of tactile pavement image segmentation algorithm.Based on the original Seg Net network model,the coding part of the Seg Net model is replaced with the feature extraction part of Mobile Net V3.The experimental results show that the improved model can segment the tactile pavement area more accurately and improve the image segmentation speed by training on the homemade dataset,which meets the requirement of real-time for the guide system.