Mobile service robot of much reality application is becoming an active topic in robot °ied. Navigation and localization are the key and fundament for service robot technology. Multi-sensor system combining vision and other sensors is desirable to improve robot localization such as accuracy,real-time and robustness, and of great theory and application value for the improvement and development of service robot. Therefore, multi-sensor based robot navigation and localization have attracted more attention in mobile service robot field in the past years. In this thesis, multi-sensor based localization technology is explored for mobile service robot in unstructured environment. Some relevant topics such as vision detection and feature matching, active vision modelling, multi-sensor calibration, multi-sensor based localization, and vision-guided approaching and grasping are discussed in this thesis. The main contributions of this thesis include following issues: 1. A modified SIFT algorithm is proposed to extract feature and find correspondence in real environment. In this algorithm, state tag and color constrain are introduced to speed the feature extraction and matching, and an epipolar constrain based pseudo algorithm is developed to improve matching accuracy. Meanwhile, a classified Hough Transform is proposed to improve the effectiveness and accuracy of line extraction for rectangular mark. 2. An extrinsic parameter calibration method is proposed for a kind of active stereovision system with independent-rotation cameras to improve convenient of visual system measurement. 3. Motion-based self calibration approach is proposed for the multi-sensor system with odometer, inertial measurement unit and active vision platform in unstructured environment, and make the system calibration more flexible and convenient. 4. Multi-sensor based localization algorithm is developed for mobile service robot. The Kalman filter is introduced to fuse the data captured by odometry and Inertial measured unit and to improve localization's robust. With the correspondences in series images, maximum likelihood motion estimation algorithm is developed to estimate robot's position as well as those features'. The epipola constrain is introduced to speed prediction and track of visual features. 5. A new visual measurement model is developed for robot approaching and grasping. A more flexible translation based calibration approach using single point is proposed for this visual model. A...
修改评论