CAJ | 학술논문

鉴于室外自主移动机器人视觉导航的应用需求，提出一种提高非结构化道路跟踪准确性和鲁棒性的算法。首先，利用局部二值模式LBP的纹理特征和颜色特征建立道路区域的H—S—LBP特征模型，利用特征模型对视频序列图像进行反向投影，将图像分割为道路区域和背景区域。同时，采用卡尔曼滤波器更新道路特征模型。通过反向投影与特征模型更新的交替完成，实现非结构化道路跟踪。实际应用表明，提出的算法能够准确跟踪视频图像中道路区域，对复杂多变的室外环境适应性较强．道路边缘处区分度较高，满足实时性要求。
감우실외자주이동궤기인시각도항적응용수구，제출일충제고비결구화도로근종준학성화로봉성적산법。수선，이용국부이치모식LBP적문리특정화안색특정건립도로구역적H—S—LBP특정모형，이용특정모형대시빈서렬도상진행반향투영，장도상분할위도로구역화배경구역。동시，채용잡이만려파기경신도로특정모형。통과반향투영여특정모형경신적교체완성，실현비결구화도로근종。실제응용표명，제출적산법능구준학근종시빈도상중도로구역，대복잡다변적실외배경괄응성교강．도로변연처구분도교고，만족실시성요구。
An unstructured road tracking algorithm based on feature model is presented in view of the application requirement in vision-based navigation of autonomous mobile outdoor robot. It aims at improving the accuracy and robustness of tracking results. First of all, H-S-LBP feature model of road is built based on local binary pattern and color. Video sequences are prnjected by feature model which is turned into back-projection images, and they are segmented into road and background. Meanwhile, the feature model of road is updated by Kalman filtering. The system alternately executes back projecting and feature model updating in order to achieve unstructured road tracking. Applications show that the approach can track unstructured road precisely and is more adaptable to the environment. The discrimination of road edge is good. It also can be applied to practice in real time.