针对无人驾驶车环境感知技术,基于D-S证据理论融合多传感器信息,旨在解决障碍物身份识别技术难点.基于CCD和激光传感器建立信息融合系统,并提取每种障碍物的5个特征证据,包括距离对比度特征、平行四边形特征、边缘形状特征、灰度纹理特征和颜色特征.再根据目标类型和环境加权系数选择经验公式,通过模糊插值法求取身份隶属度近似获得各特征对目标的相关系数构造基本概率赋值函数.最后制定Dempster组合规则,融合多传感器特征信息识别障碍身份.试验表明本文方法能够准确有效地获取基本概率赋值函数,D-S证据理论融合方法提高了障碍物身份识别的准确性和鲁棒性.%Autonomous navigation in cross-country environments presents many new challenges including obstacle perception for unmanned ground vehicle. A new method suitable for recognizing obstacle is proposed. The first step is to build the sensor fusion system by using sensors such as CCD and ladar, then to extract five different types of features, including distance contrast, parallelogram rate, edge-shape-factor, gray texture and HSV value. The experiment formula is selected according to the types of obstacle and weight efficiency to calculate basic probability assignment (BPA). The subordinatien to each event in identification framework is obtained by using the fuzzy interpolation. It is supposed that the subordination is equal to correlation coefficient in the formula. Finally, dempster rules are used to integrate sensors information and the obstacle is recognized based on the D-S theory of evidence. The test results indicate that the resolution of BPA is correct, thus improving the validity and robustness of cross-country environment perception based on the new method.
展开▼
