Road Sign Recognition System Based on HOG Descriptors and GentleBoost with Sharing Features

• Main Authors: Chun-HaoChang, 張峻豪
• Other Authors: Jenn-Jier Lien
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/08339302826893952101

碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 98 === Nowadays, the number of vehicles is growing rapidly, and more and more intelligent transportation systems are developed for assisting drivers. Road sign detection and recognition is extremely important for safe and careful driving, this system can not only inform the driver about the condition of the roadway but also support the driver during the tedious task of remembering the large number of road signs. In this thesis, we propose a fast road sign detection and recognition system. This system takes advantage of the HSI color space to filter most of the false alarms. Histogram of Oriented Gradient (HOG) is then used for the shape detection (including circular, triangular, rectangular, and octagonal signs), finally, the candidate blobs that pass through the shape detection is recognized by a GentleBoost detector and rotation, scale, translation-invariant (RST-invariant) template matching. In detection step of our algorithm, deformation, scale, and partial occlusion problems can be solved by utilizing HOG information; for recognition step, Color information is used for training GentleBoost detector to ensure the accuracy of the system; and the achromatic part of the candidates are matched to the templates by RST-invariant template matching. The main advantage of this system is that it can detect and recognize road signs efficiently and accurately.