Feature recognition and obstacle detection for drive assistance in indoor environments
| dc.contributor.author | Zheng, Chunhui | |
| dc.date.accessioned | 2022-05-05T02:53:32Z | |
| dc.date.available | 2022-05-05T02:53:32Z | |
| dc.date.issued | 2012 | en |
| dc.description.abstract | The goal of this research project was to develop a robust feature recognition and obstacle detection method for smart wheelchair navigation in indoor environments. As two types of depth sensors were employed, two different methods were proposed and implemented in this thesis. The two methods combined information of colour, edge, depth and motion to detect obstacles, compute movements and recognize indoor room features. The first method was based on a stereo vision sensor and started with optimizing the noisy disparity images, then, RANSAC was used to estimate the ground plane, followed by a watershed based image segmentation algorithm for ground pixel classification. Meanwhile, a novel algorithm named a standard deviation ridge straight line detector was performed to extract straight lines from the RGB images. The algorithm is able to provide more useful information than using the Canny edge detector and the Hough Transform. Then, the novel drop-off detection and stairs-up detection algorithms based on the proposed straight line detector were carried out. Moreover, the camera movements were calculated by optical flow. The second method was based on a structured light sensor. After RANSAC ground plane estimation, morphology operations were applied to smooth the ground surface area. Then, an obstacle detection algorithm was carried out to create a top-down map of the ground plane using inverse perspective mapping and segment obstacles using a region growing-based algorithm. Both the drop-off and open door detection algorithms employ the straight lines extracted from depth discontinuities maps. The performance and accuracy of the two proposed methods were evaluated. Results show that the ground plane classification using the first method achieved 98.58% true positives, and the figure improved with the second method to 99%. The drop-off detection algorithms using the first method also achieved good results, with no false negatives found in the test video sequences. The system provided the top-down maps of the surroundings to detect and segment obstacles correctly. Overall, the results showing accurate distances to various detected indoor features and obstacles, suggests that this proposed colour/edge/motion/depth approach would be useful as a navigation aid through doorways and hallways. | en |
| dc.identifier.uri | https://hdl.handle.net/10092/103631 | |
| dc.identifier.uri | http://dx.doi.org/10.26021/12731 | |
| dc.language | English | |
| dc.language.iso | en | en |
| dc.rights | All Right Reserved | en |
| dc.rights.uri | https://canterbury.libguides.com/rights/theses | en |
| dc.title | Feature recognition and obstacle detection for drive assistance in indoor environments | en |
| dc.type | Theses / Dissertations | en |
| thesis.degree.discipline | Computer Science and Software Engineering | |
| thesis.degree.grantor | University of Canterbury | en |
| thesis.degree.level | Masters | en |
| thesis.degree.name | Master of Science | en |
| uc.bibnumber | 3144561 | |
| uc.college | Faculty of Engineering | en |