On-Road Vehicle and Lane Detection

We implement lane detection using edge detection, Hough transforms, and vanishing point filtering in Hough space; the car detection is implemented by using histogram of oriented gradients feature descriptors and classified by linear support vector machines. Hard-negative mining is applied to alleviate detection of false positives; with the information of vanishing point along with prior knowledge such as the width of the lanes, we reconstruct the 3D ground plane and estimate the distance from the camera to the cars in the front from monocular vision. Keywords—Car detection, Lane detection, Hough transform, HOG, SVM, hard negative mining.