Multi geometrical image processing based on active vision agent

Abstract

This paper presents an image processing capable of detecting basic shapes that communicate as intelligent agent by active vision sensor system. We also propose image recognition algorithm to facilitate a binocular camera to be able to detect and recognize multi geometrical object shapes. We investigate within plain and crowded background environment, indoor and outdoor along with low noise and very high noise that complicate the edge shape. The evaluation result shows plain environment give the highest recognition percentage followed by indoor and outdoor. Meanwhile the unsatisfied result is due to the high noises that make the gradient magnitude cannot distinguish between the multi geometrical edges and background because of their similar or shared pixels. However, the image processing efficiency is improved and able to distinguish image in real time navigation and amenable to spatial agent memory architectures. In near future, the result shall be used as a capstone for future application and development in humanoid robot sensor field.