Positioning algorithm for agricultural machinery omnidirectional vision positioning system

Abstract: The absolute position of the sensor with respect to identifying the coordinate system is estimated based on identifying the azimuth angle for agricultural machinery full visual positioning system. The positioning algorithm was a key for the omnidirectional vision positioning system. A novel agricultural machinery positioning system was developed based on an omnidirectional vision and four or less than four artificial landmarks utilizing abundant imaging information and unchangeable directional angle for spatial point and imaging point. The positioning system has preferable features for agricultural machinery compared with GPS because of the simpler structures, higher precision, better adaptability and able to work in the night. Four artificial landmarks are built on the four corners of the enclosing rectangle around the working area. The estimated position is calculated according to the circumferential theorem and geometric transformation based on the direction angles of the landmarks with the principal point of camera. The algorithms mainly included the imaging system calibration, noise elimination, landmarks' features extraction and position estimation. Considering the noise of environment or some light obstacles, one or more of four landmarks' features extractions may be defeated. The algorithms of four landmarks detection and three landmarks detection to estimate sensor position were studied. First, the coloured landmark pixels beyond the threshold are extracted as a small area and the center of gravity is calculated for the extracted small area representing the position of one landmark. Then, the position of four representative landmarks is obtained and the four directional angles of the landmarks are estimated using only one omnidirectional image. Sensor position is able to be estimated using the center of gravity of the four intersections formed by four arcs according to geometric transformation based on the four directional angles of the extracted four landmarks. And the sensor position is also able to be estimated using the center of gravity of the three intersections formed by three arcs according to geometric transformation based on the three directional angles of the extracted three landmarks. Pointing test and camera tilt test were conducted on the level ground in an area of 30 m × 30 m outside under natural sunlight. 25 points every 5 m in the x and y axes were selected in the pointing test, and 7 points (x, y) = (0, 15), (5, 15), (10, 15), (15, 15), (20, 15), (25, 15), (30, 15) in the middle of the square area were selected in the camera tilt test. Chuo Seiki precision equipment for adjusting the camera angle was operated by hand. Pointing test results showed that the maximum absolute distance error was 25.72 cm; the average absolute distance error and RMS distance error of sensor position were 14.06 and 14.75 cm, respectively. Sensor tilt test results showed that the positioning accuracy was influenced by the tilt angle. If the tilt angle more than 5°, the influence was obvious. In conclusion, the proposed algorithm is effective and simple, and the program speed is rapid. In practice, it is necessary to consider the compensation for positioning error via sensor tilt when using under the condition of tilt angle greater than 5 degree or rugged environment. This system is a potential substitute or compensation for GPS in agricultural machinery navigation required for indoor and outdoor environments in the future.