融合多传感器信息的移动机器人自定位方法

Mobile robot self-localization based-on multi-sensory information fusion

提出一种给定环境模型下移动机器人全局自定位算法 ,通过融合声纳传感器和视觉传感器的异质传感信息把具有多模态、鲁棒性强的Markov方法和单模态、高效准确的EKF方法组合应用并加以改进 ,来实现准确和快速的全局定位 ,同时提高位姿跟踪的准确性 .Markov方法中位姿空间的低分辨率离散减小了存储需求 ,声纳感知模型对位姿空间分布进行初始化并提供了全局的位姿假设 ,视觉感知模型实现了位姿分布更新 ,而基于视觉特征的EKF方法则提高了定位的精度 .实验结果验证了本方法的有效性 .

A computation method is proposed for global localization which utilizes information from both sonar and vision sensors. The method combines multimodal, robust Markov and unimodal, efficient extended Kalman filter (EKF) localization with significant improvements for global localization and position tracking. In Markov localization, memory requirements are reduced with low-resolution discretization of pose space. The pose space distribution is initialized and pose hypotheses are acquired through sonar sensor model, and sensor update is accomplished through visual sensor model. Then vision-based EKF is utilized for localization precision. Experimental results demonstrate the validity of the approach.