Zebra-Crossing Detection and Recognition Based on Flood Fill Operation and Uniform Local Binary Pattern

Detection of zebra-crossing region from an image is an important and demanding task to support visually impaired people to navigate the street crossing safely in the exterior environments. In this paper, a zebra crossing detection and recognition method is presented where the crossing region is detected by employing the image processing techniques such as adaptive histogram equalization, flood fill operation, and Hough transforms and is recognized through the uniform local binary pattern with support vector machine (SVM) classifier. For that, the contrast and sharpness of the zebra crossing image is improved by the adaptive histogram equalization if the image's intensity value is less than an empirical threshold value. After that, the preprocessed zebra-crossing image is converted to the binary image by using the Otsu's method. Furthermore, the morphological and flood fill operations are applied to the binary image to extract the largest candidate object. The edges of the largest candidate object are detected by utilizing the canny operator. From the edges, the potential longest horizontal edges are estimated by eliminating the vertical edges using four connected method and filtering the small edges using statistical threshold procedure. Finally, the potential parallel horizontal edges are justified as zebra-crossing edge lines by drawing the Hough lines and detect the zebra-crossing region of interest (ROI). Then, the SVM classifier is applied to the detected ROI region to recognize the zebra-crossing region where, rotational invariant uniform local binary pattern is utilized to extract the features of candidate region. The results of the simulation show that the proposed method effectively detects and recognizes zebra crossing regions from various zebra-crossing images. Moreover, it shows superior performance than the state-of-the art methods in terms of recognition.