Development of a vision based object classification system for an industrial robotic manipulator

Due to the growing cost of raw materials, workmanship, energy, and growing competition environment, manufacturers are forced to produce cheaper and higher quality productions. This force results in the need for automation techniques. We have developed a vision based system for industrial robotic manipulators to classify objects on a moving conveyor. Low and intermediate level image processing algorithms are implemented as a first step. In the implementation of application level image processing, a neural network is designed to classify objects, and moment invariants are used as feature vector set. The system is tested for different objects moving on a conveyor. The speed and the position of the object are computed. This information represents the most used criteria in the control of robots. The hardware implementation and overview of the algorithms with their results are presented.