A novel lane feature extraction algorithm implemented on the TMS320DM6437 DSP platform

Abstract

This paper presents an implementation of a novel lane edge feature extraction algorithm based on digital interpolation. The method is based on the observation that by zooming towards the vanishing point, the positions and the sizes of the lanes will not change significantly while other objects will. Comparing the zoomed image with the original image allows us to remove most of the unwanted features from the lane feature map. The proposed algorithm shows outstanding performance on extracting features belonging solely to the lanes embedded in severe noise environment. The algorithm runs in real-time on the TMS320DM6437 DSP platform from Texas Instruments. The system implemented can achieve at least 23 fps, without performing any code optimisation but only considering memory management.