| dc.contributor | 地政系 | |
| dc.creator (作者) | Lin, Jou An;Chio, Shih-Hong | |
| dc.creator (作者) | 邱式鴻 | zh_TW |
| dc.date (日期) | 2013-10 | |
| dc.date.accessioned | 26-May-2015 18:07:40 (UTC+8) | - |
| dc.date.available | 26-May-2015 18:07:40 (UTC+8) | - |
| dc.date.issued (上傳時間) | 26-May-2015 18:07:40 (UTC+8) | - |
| dc.identifier.uri (URI) | http://nccur.lib.nccu.edu.tw/handle/140.119/75311 | - |
| dc.description.abstract (摘要) | To automatically recognize traffic signs from vehicle-borne images can be used for the applications in road safety, the maintenance of reverent road facilities, navigation for cars and pedestrians, even for the development of future driver assistance system. From the literatures, two approaches are employed to recognize traffic signs from images. One approach is to detect them by hue and saturation of traffic sign images, the other is to recognize them by the descriptors of traffic sign images that are extracted by using algorithms, e.g. SIFT algorithm (Lowe, 2004). However, neither of them recognizes successfully when the vehicle-borne images had large geometric distortion or was affected by complex background, weather, shadow, and illumination. The foregoing problems can be improved by capturing images carefully or image processing except for large geometric distortion. A-SURF (Affine Speeded up Robust Features) algorithm was developed by Pang (2012). This algorithm is affine invariant and computation efficient, therefore, this study will try to employ A-SURF algorithm to overcome the problem of large geometric distortion. Firstly, the descriptor database of standard traffic sign images collected from the ministry of transportation in Taiwan is built. After that, the nearest distance between each keypoint descriptor in each standard traffic sign image in database and every keypoint descriptor in the traffic sign image extracted from vehicle-borne image will be determined by k-d tree. After finding all the best corresponding keypoints between the standard traffic sign image and the extracted traffic sign image. Wrong keypoint correspondences should be removed by using RANSC algorithm (Fischler, 1981) on the assumption of perspective transformation between standard traffic sign image and the extracted traffic sign image. Finally, the successful recognition is determined by the quality of affine transform between the keypoint correspondences. From the tests, the results of our proposed approach and the relevant problems will be also discussed. | |
| dc.format.extent | 176 bytes | - |
| dc.format.mimetype | text/html | - |
| dc.relation (關聯) | 34th Asian Conference on Remote Sensing 2013, ACRS 2013, 2, 2013, 1270-1276, 34th Asian Conference on Remote Sensing 2013, ACRS 2013; Bali; Indonesia; 20 October 2013 到 24 October 2013; 代碼 105869 | |
| dc.subject (關鍵詞) | Algorithms; Automobile drivers; Computational efficiency; Geometry; Remote sensing; Traffic signs; Trees (mathematics); Vehicles; Affine SURF; Automatic recognition; Future driver assistance systems; Geometric distortion; Keypoint correspondences; Perspective transformation; Speeded up robust features; Vehicle-borne; Image processing | |
| dc.title (題名) | Automatic recognition of traffic signs from vehicle-borne images | |
| dc.type (資料類型) | conference | en |
