基于D-S证据理论的障碍目标身份识别

Obstacle identification based on dempster-shafer theory of evidence

以越野环境中典型的障碍物为识别目标,选用单目视觉与激光扫描仪建立融合系统,基于D-S证据理论融合多传感器信息,实现UGV对障碍目标的身份识别。首先将每个传感器的观测数据从观测空间变换到证据空间,对每种身份分配一个基本概率赋值;融合系统再根据Dempster组合规则计算各个命题组合后的概率赋值函数和相应的信任度区间,然后计算综合概率赋值函数和信任度区间;最后根据计算结果和决策规则进行障碍身份识别。试验表明:该方法优于利用单个特征识别障碍物身份,能大大提高系统对于障碍物的识别分类能力。

Sensor data fusion system is composed from single color CCD and laser scanner. Based on Dempster-Shafer theory of evidence different sensor data are fused together to classify three kinds of typical obstacles. Firstly, sensor data are transformed to evidence which attributes to the obstacle＇s characteristic. Since identity of the current obstacle is unknown, basic probability assignment is assigned to every classification. Then Dempster fusion rules are used to compute the ultimate BPA and Believe interval. Finally, obstacle identity classification is defined according to the identification rules. Test results show that D-S theory is superior to single characteristic method when classifying the obstacle identification.