An Augmented Sliding Window Technique to Improve Detection of Curved Lanes in Autonomous Vehicles

摘要

In this paper, an improved curved lane detection algorithm using multiple sliding windows is proposed. Image processing techniques like color-based thresholding; edge detection and perspective transformation are mainly used for retrieving the data about the lane points from the input image. The selection of initial sliding window is very important and hence previous lane starting points are saved in the cache to efficiently detect the moving lanes. Basic sliding window approach is not suitable in sharp curves and dashed lanes. Multiple sliding window technique is able to detect the sharp curves and dashed lanes. Once the left and right lanes are detected, the center of the left lane and right lane is found; vehicle's deviation from actual center of the lane is calculated. The performance of the proposed multiple sliding window technique is compared with the basic sliding window approach. For a data set of 938 images, the detection accuracy of the lanes using the proposed algorithm is observed to be 96.26%.