基于贝叶斯估计理论的自主机器人双目视觉相机量化误差数据最优融合方法研究

项目名称： 基于贝叶斯估计理论的自主机器人双目视觉相机量化误差数据最优融合方法研究

项目编号： No.61304236

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 王小刚

作者单位： 哈尔滨工业大学

项目金额： 23万元

中文摘要： 本项目针对自主机器人具有量化误差特性的双目视觉相机测量数据与里程计信息的最优融合问题展开研究。传统卡尔曼滤波融合传感器数据要求量测噪声为高斯白噪声，这与双目视觉相机测量数据的量化误差特性相矛盾，必然导致滤波精度下降，尤其是当双目视觉相机与所观测的特征点距离较远时，量化误差的影响更为显著，这时甚至会出现滤波发散的现象。为解决上述问题，首先深入分析双目视觉相机量化误差的产生机理，提出量化误差建模方法，进而推导出量化误差的随机模型；然后，以量化误差的建模成果为基础，结合粒子滤波和基于求解Fokker-Planck方程的非线性滤波的改进研究成果，解决双目视觉相机测量数据与里程计信息的最优数据融合问题。最后，设计和搭建导航系统硬件试验平台，对比分析各种量化误差随机模型和数据融合算法的性能。对双目视觉相机测量数据量化误差特性以及融合问题的研究有助于提高自主机器人在导航、控制、目标跟踪等各方面能力。

中文关键词： 双目视觉相机；量化误差；量化数据融合；粒子滤波；Fokker-Planck方程

英文摘要： The optimal fusion between stereo vision camera data with quantization error and dead-reckoning is investigation. The conventional kalman filtering demands that the measurement noise is Gaussian noise, which is quantization error for stereo vision camera. This leads to the accuracy decrease of kalman filtering. Especially, when the distance between the stereo vision camera and the features is far, the effect of quantization error is obvious. The divergence of filtering happens some time. To solve these problems, the reason for quantization error is analyzed firstly. And the modeling method for quantization error is proposed. As a result, the stochastic model of quantization error is derived. Then, the particle filtering and nonlinear filtering based Fokker-Planck equation are incorporated with the stochastic model of quantization error to solve the problem of fusing the data of stereo vision camera and dead-reckoning. Finally, the hardware platform of integrated navigation system is constructed. The stochastic models of quantization and filtering algorithms are compared. The investigation of quantization error and its fusion helps to improve the ability to navigate, control and tracking target for autonomous robot.

英文关键词： Binocular Vision Camera；Quantization Error；Quantized Data Fusion；Particle Filter；Fokker-Planck Equation