This paper proposes a cooperative visual object tracking by multi-robot system, where robust cognitive sharing is essential between the robots. However, one of the main issues in vision-based distributed observation is the significant differences in the background image for the interested object. According to the observing point of the robot, effective invariant feature to identify the interested object is different. In this paper, we propose an ambiguity index to select better feature algorithm for object tracking. Experimental result shows promising result for the effective multi-robot cognitive sharing.