文摘
The detection of the ball when it is not on the ground is an important research line within the Middle Size League of RoboCup. A correct detection of airborne balls is particularly important for goal keepers, since shots to goal are usually made that way. To tackle this problem on the CAMBADA team , we installed a perspective camera on the robot. This paper presents an analysis of the scenario and assumptions about the use of a single perspective camera for the purpose of 3D ball perception. The algorithm is based on physical properties of the perspective vision system and an heuristic that relates the size and position of the ball detected in the image and its position in the space relative to the camera. Regarding the ball detection, we attempt an approach based on a hybrid process of color segmentation to select regions of interest and statistical analysis of a global shape context histogram. This analysis attempts to classify the candidates as round or not round. Preliminary results are presented regarding the ball detection approach that confirms its effectiveness in uncontrolled environments. Moreover, experimental results are also presented for the ball position estimation and a sensor fusion proposal is described to merge the information of the ball into the worldstate of the robot.