Development of an Embedded Artificial Vision System for an Autonomous Robot

This paper presents a completely autonomous system which is able to find, detect and recognize objects by different patterns. A low-cost artificial vision system was implemented using only a webcam and a complex image processing algorithm, SIFT. This vision system can be embedded in any autonomous robot, but for experiments and demonstration it was tested on the Khepera III autonomous robot platform and the extension board KoreBotLE with ARM PXA255 XScale 400MHz processor. This paper describes all the steps and details necessary for implementing embedded artificial vision at all levels: hardware, firmware, software and behaviour. There are given solutions for different problems caused by hardware restrictions and then the results and constraints of this low-cost vision system are presented. The software system was developed to overcome hardware limitation, so that the robot could gather visual data simultaneously to the image processing, in real-time. The main contribution of the paper is the development of a method of using a complex and highly efficient image processing algorithm, SIFT, on an embedded system with reduced processing capabilities. A number of experiments were done to prove that this low-cost artificial vision system is feasible and enhances the robot’s cognitive skills by processing complex visual data obtained from a simple webcam.