Landmark guided trajectory of an automated guided vehicle using omnidirectional vision

摘要

The omnidirectional camera is very useful in tracking a landmark for automated guided vehicle (AGV). The omnidirectional camera can sense object 360° around the AGV thus eliminating the need of camera panning or robotic reorientation. The image produced by the omnidirectional camera is usually highly distorted. However, one feature of the image captured by an omnidirectional camera is that the distortion only against the height of the object. Object with negligible height has negligible image distortion. With this feature in mind, this research investigates the trajectory generated from an AGV towards an identified and recognized landmark using omnidirectional camera without rectifying the distortion into perspective view. The research work involves landmark identification and recognition using image processing step. The landmark used, was enlarged to four different sizes, code-128 barcodes with cyan background and red orientation marker. The landmark identification and recognition is processed from the image captured by the omnidirectional camera. The camera was mounted on the AGV and remain as the sole range sensor for the AGV to sense its environment. Three fundamental trajectories used in robotics navigation namely straight, left turn, and right turn were experimented to present the trajectory of an AGV guided by a landmark. The AGV was modelled using Bicycle Model. The trajectory of the AGV is then simulated using MATLAB/Simulink. Next, the simulation work is validated with the experimental work. A proportional control is applied in the experimental work for the AGV move toward the landmark. All experiments were conducted in a laboratory environment with controlled illumination. The work thus demonstrate that the image captured using omnidirectional camera can be used to identify and recognize a landmark without going through any typical omnidirectional image unwarping process into a perspective view. The important navigational information for the vision-based-AGV can be extracted directly from the camera feed.