Object recognition using multidirectional vision system on soccer robot

An autonomous mobile robot should be equipped by a sensory system for sensing the environment before making any decision. In soccer robots, the sensory system uses cameras as the vision system device to recognise the objects on its working environment. The objects to be considered are the ball, goalpost, and the marks on the field. The vision system of a soccer robot usually has its limitations while capturing object due to the camera specification. This paper discusses a method to recognise object in a better way by using multiple camera for detecting object in different direction and distance. The distant objects can be detected by using a camera which is focused on a straight line, and able to detect capture object in long distance. Otherwise, the close objects can be detected by using a camera which is focused on nearby location, yet able to capture multiple direction in the same time to broaden the ability to detect certain objects. Digital image processing techniques such as image thresholding, Gaussian blur, Canny edge detector and Hough transform are implemented using OpenCV to process image captured by the cameras. The information then forwarded to the robot coordination system to make decision. Testing has been done to prove that the vision system is able to extract information about the objects on the field. Overall, the vision system using cameras is able to extract the information needed by soccer robots to recognise the environment. Integration testing of the whole soccer robot system shows that the robot is able to recognise, approach, then herd and kick the ball toward the goal.