Fusing Optimal Odometry Calibration and Partial Visual Odometry via A Particle Filter for Autonomous Vehicles Navigation

Abstract— Autonomous vehicles are increasingly becoming ubiquitous in the 21st century; they find application in agriculture, industry, airplanes, cars, service robotics, and others; in order to display autonomous guidance, a vehicle needs to estimate its position and orientation relative to an arbitrary coordinate system; to do so, several sources of information can be used, including images, global positioning systems, inertial measurements or odometry, each according to the application; methods, such as Kalman Filter can be used to combine the several sources of information; however, the more accurate each source of information is, the better the estimation of vehicle position and orientation will be; therefore, the calibration of the parameters of the odometrical systems in autonomous terrestrial vehicles is a must; visual guidance is also an important technology used for vehicle guidance. In this paper, it is presented an off-line method for odometry calibration using a genetic algorithm and the fusion of odometry data with heading information from camera data; a particle filter is used to fuse the data from the optical encoder and the camera. This method was tested in an Automated Guided Vehicle (AGV) with tricycle topology, demonstrating high accuracy in position estimation and guidance through arbitrary paths.