Research On Vision Based Odometer For Indoor Mobile Robot

The problem of the self-localization of indoor mobile robot is a hot research direction. Under the background, the odometer vision-based for indoor mobile robot is researched, aimed at filling up the deficiency of traditional sensor in self-localization of robots. To the encoding disk, it didn't need to record the running times to get the robot's position which would be influenced by the skidding of the wheel,and to the laser sensor and other, the camera is a passive transducer, which didn't need to transmit signals and receive them.To solve the problem of self-localization of robots, this dissertation makes a study on visual odometer and it includes the following works:To get higher precision and reduce the effect of the niose, the way of working of the odometer is choosed. To reasonably select the hardware and the algorithm, the index of the odometer is analysised. Then the hardware scheme and the software process are proposed.The performance of the key algorithm of the extraction of feature points-SURF and the nearst neighbor matching algorithm is analyzed and tested. According to the analysis of the model of indoor mobile robot and odometer, the algorithm of reasonably cutting of the pictures is proposed.As a result, the time is reduced and the speed of processing is improved.The camera model is choosed and the camera is calibrated. Then thepoints under the frame coordinate are converted to the other ones under the camera coordinate. The distortion correction and the slant correction is made. At last, the adjacent frames'transformation matrix under the camera coordinate is got by RANSAC algorithm.The robot's position and attitude are got under the robot origination coordinate. Then according to the calibration experiments of the rotational motion and the translational motion, the transformation matrix between the robot origination coordinate and the camera origination coordinate is acquired, after which the requirement of the installation of the camera is reduced and the accuracy of the system is improved.The off-line experiments are designed: The robot is directed to move along a rectangle, a circle without autorotation and a circle. The precision of the odometer was verified by the destination coordinates, the chainage and the attitude. And then the on-line experiments are taken: The robot is directed to move along a line several times, move with only autorotation and move along a circle. The precision and the real-time capability of the odemetry are verified by the destination coordinates, the chainage and the attitude.This dissertation preliminary meets the requirements of the odometer design, and establishes the foundation for further research on visual odometer.