Object Recognition in Robot Football Using a one Dimensional Image

Robot football is a challenging task in the world of autonomous robots and related research areas. One of the main problems in this task is to determine the presence and the location of objects from the robot’s point of view. This paper introduces methods to recognize the objects in a football field, by the aid of vision. The methods are tested on a simulator [1] of a Khepera [2] autonomous mobile robot with an onboard one dimensional, 256 gray level black-white camera. The objects in the field are the light gray walls, a simulated yellow tennis ball, and a black goal area. The specifications of the robot, and the field, are determined by the FIRA association [3]. Three methods are introduced in this work. In the threshold based method, a high pass filter is used multiple times to detect the rising and falling edges, then a threshold is applied over the detected edges to eliminate the false edges. After the detection of the edges, regions are detected. By using the number of regions, the mean value of each region, and the standard deviation of the regions, several rules, which are needed to classify the regions, are extracted. These rules use the relations between the objects, relative to each other. The second method is based on a neural network, and the last method uses fuzzy expert systems, to recognize the objects.