Computer stereo vision technique has been prevalently used in various automotive applications for depth perception. This thesis mainly focuses on developing a computationally efficient and highly accurate disparity estimation algorithm for three automotive applications, i.e., lane detection, road surface 3-D reconstruction and pothole detection. Firstly, the real-time implementation of an efficient stereo matching algorithm is proposed to acquire dense disparity maps for road scenes, where the disparities are estimated iteratively whereby the search range on each row is propagated from three estimated neighbouring disparities on the lower row. The dense vanishing point estimation in a multiple lane detection system is then improved using the obtained disparity information, where the author uses RANSAC to update the road model parameters. The proposed lane detection algorithm is implemented on a heterogeneous system for real-time purposes. Furthermore, the disparity estimation algorithm is used to reconstruct the 3-D road surface. This is achieved by first transforming the perspective view of the target frame into the reference view. To achieve the millimetre accuracy required for road condition assessment, a disparity map with subpixel resolution is achieved by performing a parabola interpolation enhancement. Moreover, a novel disparity global refinement approach developed from MRF and FBS is introduced to further improve the accuracy of the estimated disparity map. Finally, the estimated disparity maps and reconstructed 3-D point clouds are used in a pothole detection system. The disparity map is first transformed to better distinguish potholes from the road surface. To achieve a higher processing efficiency of disparity map transformation, GSS and DP are utilised to estimate the transformation parameters. Then, the disparity map is modelled as a quadratic surface and the gradient information is also integrated into the process of disparity modelling. By comparing the difference between the actual and fitted disparity maps, the potholes can be detected.
|Date of Award||25 Sep 2018|
- The University of Bristol
|Supervisor||John G Rarity (Supervisor) & Naim Dahnoun (Supervisor)|