Shadow Lane Robust Detection by Image Signal Local Reconstruction

Abstract

In order to resolve shadow interference and slow image processing speed in the visual navigation of unmanned vehicles on city roads, a new lane detection algorithm based on the Inverse Perspective Mapping(IPM) of vertical sub-image reconstruction using local bands was proposed. The lanes IPM aerial image was obtained using the projective transformation of three-dimensional images of city roads. The ROI portion of the IPM map was decomposed and analyzed using the sym3 wavelet, and after analyzing and comparing the experimental results the first and second levels of the vertical sub images were selected for the reconstruction, compression, and removal of shadows from the original image. The Canny algorithm was proposed and adopted in order to extract the edge features of the reconstructed images according to real-time road image quality. A modified, polar angle, constraint-based, fast Hough transform was used to locate the candidate lanes. Finally, the main control point, straight-line-fitting algorithm was used to fit the final lanes, which achieved the precise location and recognition of the lane lines. The California Polytechnic lane data set, a public data platform, which is now widely used in road visual recognition fields around the world, was selected for the testing and verification of the algorithm. The results of the experiment and actual operation indicated that, despite having a storage space of approximately one tenth the size of similar algorithms, this detection algorithm effectively solves the issue of shadow removal, meets the real-time requirements of roadway image processing systems for unmanned vehicles with a recognition time of less than 20ms, and is robust. 2016 SERSC.