Object detection is a broad area that comprises a variety of methods for measuring and categorising objects, through the use of various sensor systems. These sensor systems differ according to their environments, allowed lighting conditions and the type of objects that they can detect; they have advantages and disadvantages accordingly.
This diploma thesis is about object detecion in the area of robotics, especially about its use in Robot-soccer - which is increasing in popularity due to the associated research possibilities. The designed robot is required to follow the rules of the Mirosot Category, specified by the international organisation for advancement of Robot-soccer called FIRA. A digital camera is the sensor used for object detection: Starting from constraints in the measurements (based on the Mirosot Category and Rules), this thesis presents an appropriate hardware-platform and the reasons for choosing particular components. Based on standard methods of image processing and approximations for their algorithmic complexity, a software concept was sought-after, that meets the requirements. The main aim is the detection of the ball (an orange golfball) and the coloured goalposts. Since standard methods are not optimized for the specific hardware, and the efficiency is a major concern, a concept is presented, which allows basic geometric shapes (e.g. circles, rectangles) to be detected and measured. This new concept partly uses well-known methods from the standard literature, that were optimized for the required tasks; but also includes a new approach that makes it possible to handle the detection-complexity of the dynamic motion of the target objects.
The result is a image processing system with a detection-rate of 60 images per second. The precision of the ball-detection lies at +/- 1 mm at a mean distance of 50 cm. The average power consumption is 1,5 Watt and the hardware-dimensions are such, that the hardware could be housed in a cube with sides of 7.5cm length.