Multiple Lane Detection Algorithm Based on Novel Dense Vanishing Point Estimation

Pothole Detection Based on Disparity Transformation and Road Surface Modeling

“…In recent years, many authors have turned their focus towards estimating the roll angle from disparity maps [21], [28]- [31]. For example, the road surface is assumed to be a horizontal ground plane, and an effective roll angle estimation algorithm was proposed based on v-disparity map analysis [28], [29].…”

“…13; 3 compute E 0 min (θ 2 ) using Eq. 13; 4 while θ 2 − θ 1 > ε θ do 5 set θ 3 and θ 4 to κθ 1 + (1 − κ)θ 2 and κθ 2 + (1 − κ)θ 1 , respectively; 2) α Estimation and Disparity Transformation: In this subsection, we utilize DP [21] to extract the road disparity projection model from the y-disparity map. For the purpose of convenient notation, the projection model is also referred to as the target path.…”

“…where m(d, y) represents the y-disparity value at the position [d, y] and λ is a positive smoothness term [21]. E 1 represents the energy of a possible target path in the y-disparity map.…”

“…A common practice in 3D modeling-based pothole detection algorithms [15], [16] is to fit a quadratic surface to either a 3D point cloud or a 2D disparity map. In parallel axis stereo vision, the point cloud is generated from the disparity map as follows [21]:…”

Real-time implementation of stereo vision based on optimised normalised cross-correlation and propagated search range on a GPU

“…For example, authors in [5] proposed a quasi-dense stereo matching algorithm named GCS which propagates the search range from a collection of estimated confidential disparities to their neighbours in order to save redundant computations in block matching as well as to improve the accuracy of the estimated disparity map. Similarly, our previous work [6] presented an efficient disparity estimation algorithm where the search range at row v is propagated from three estimated neighbouring disparities located at row v + 1, which is more suitable for various autonomous vehicle applications with its higher accuracy and lower computational complexities [7]. After that, an optimised version was presented in our previous paper [8] with the consideration of the search range suggested by a horizontal neighbouring disparity, which further enhances the disparity map precision but is difficult to implement on various parallel computing architectures for real-time purposes.…”