| As a service robot, sweeping robot can replace traditional manual cleaning work and has broad market prospects. At present, the domestic sweeping robots is cheap but low cleaning efficiency, since they have no navigation and positioning functions and lack effective path planning. On the contrary, many foreign sweeping robots with navigation and positioning fuctions show high cleaning efficiency, but the price is high, which is not conducive to popularization. In view of the present situation of the domestic sweeping robots, an indoor location algorithm based on monocular vision was proposed in this paper and visual information requried for positioning is from the ceiling. Compared to the use of expensive environmental sensing devices such as laser radar, using the camera greatly reduces costs. The main work of this paper is as follows:According to the difference of indoor ceiling height and decoration, a method of adaptively determining the ceiling height and an improved algorithm of feature point extraction were presented. Adaptive determination of the ceiling height needs two park location specific LED lights. After using image processing to get the LED lights centroid coordinates on the image, the ceiling height can be obtained by using camera pinhole imaging model combine with the structure of the robot. In the aspect of feature point extraction, FAST algorithm and Shi-Tomasi algorithm were united to get a certain number of feature points which are distributed reasonable in the image. Experimental results show that compared with the Shi-Tomasi algorithm, the FAST combined with Shi-Tomasi feature point extraction algorithm is about five times faster and has very close effect.According to the imaging model of the camera and the structure layout of the sweeping robot, a monocular vision positioning model of sweeping robot was established. Based on the model and the sweeping strategy of the robot, this paper calculated the robot’s translational motion, rotational motion and hybrid motion respectively. This paper presents an image matching algorithm which uses the hall sensor and gyroscope sensor data to narrow down the range of feature points matching. In addition, this matching algorithm uses ORB (Oriented Fast and Rotated BRIEF) features and hamming distance to determine the matching pairs. After eliminating the many to one matching pairs, we use RANSAC and other algorithms to do further matching pairs purification. Besides, the algorithm uses different image matching strategies for the different motion of the sweeping machine to provide accurate matching pairs for the location.In order to prove that the monocular vision localization algorithm proposed in this paper is effective, the verification experiment of the linear motion, rotary motion and rectangular loop motion of the sweeping machine were designed. To complete the experiment, many fuctions were realized on the Smart210 core board such as communication with STM32F4 by serial, using the V4L2 (Video 4 Linux 2) to control camera gray video capture and USB-WiFi positioning data transmission. A Qt demonstration interface was made on Ubuntu at PC virtual machine platform to show the waveform of the location data which comes from Smart210. Experimental results are:In the experiment of controlling the sweeping robot to move not more than 250cm in one direction, the distance error between the positioning results in the movement direction as well as its vertical direction and the actual measurement value is within 5 cm, the angle error does not exceed 5 degrees. When controlling the robot to rotate not more than 180 degrees, the error of the angle estimation and the actual measurement is within 3 degrees. Experimental results show that the proposed algorithm effectively implement the sweeping robot indoor positioning capabilities. |
PDF Full Text Request