Master of Engineering by Research
School of Electrical, Computer and Telecommunications Engineering
A vanishing point is where the perspective projections of a set of parallel lines converge in the image plane. Vanishing point estimation (VPE) is an essential task in many applications including image-based camera calibration, 3D reconstruction and road detection. In image-based road detection, the approaches that utilize vanishing point are proven to be more efficient and robust than other solutions. However, estimating vanishing point from a single road image has been a demanding task, due to the variant nature of road scene and the computation restraint of real time application. The common drawbacks of the current methods are the lack of robustness in different road scenes and the slow processing time.
In this thesis, different VPE methods based on texture or edge features are investigated. This thesis proposes an efficient VPE method which is robust to different road scenes, and possesses fast processing time suitable for real-time applications. The proposed VPE method including three stages: extracting local texture orientation and edge maps, identifying the VP search space, and calculating VP scores with a voting scheme. Color tensor features are employed to extract texture orientation and edge maps. Based on these extracted map, the optimized search spaces for VP candidates and voters are identified using Hough transform and neighborhood pixels. The voting score function is defined with a novel weighting method and a simple Bayesian classifier to adaptively adjust the algorithm.
Experiments and analysis of the VPE methods are performed with the Pedestrian Lane Vanish Point Estimation dataset, which is extended to more than 4000 images. The results shows that the proposed VPE method outperforms state-of-the- art methods in term of accuracy, robustness, and processing time.
Nguyen, Linh, Enhancements of Vanishing Point Estimation in Road Scenes, Master of Engineering by Research thesis, School of Electrical, Computer and Telecommunications Engineering, University of Wollongong, 2018. https://ro.uow.edu.au/theses1/281