Zebra crossing segmentation based on dilated convolutions

摘要

Zebra crossing segmentation is an important part of blind navigation. In this paper, We proposed a method based on Dilated Convolution Neural Network for accurate recognition of zebra crossing. The Dilated Convolution Neural Network is used to automatically identify all the regions that meet the characteristics of zebra crossing. After introducing residual network and Dilated Convolution, the detection effect of the model is greatly enhanced. The improved model improves the accuracy and reliability of zebra crossing recognition, and reduces the false detection rate. Through sample training and data testing, the results show that the Dilated Convolution Neural Network can recognize zebra crossing with an accuracy of 91.24%.