Real time active tracking and manipulation of components during assembly on conveyor belt presents a problem, because system response time from position analysis to execution of manipulation must be short. For that purpose machine vision algorithms which can recognize object and calculate its current position and orientation in real time are needed. That allows a robot to grab a component and carry out further manipulation. We developed a system for this purpose, which uses machine vision for detecting object (location and direction of movement) and controlling two robotic arms. The system is based on a video game Pong. Trajectory of a ball is analyzed with machine vision and then manipulated by two robots. The first robot is controlled with machine vision decision algorithm. Human controls the second robot and tries to pass the ball past the first robot. Machine vision also decides when each of the robots failed to respond correctly – the ball fell outside of playing surface. The first step was the development of machine vision algorithm for detecting ball position and transforming from pixel units to coordinate systems of both robots. Next step was development of control algorithms for the first robot, which uses machine vision ball orientation/trajectory data for input. We developed and compared two control algorithms. The first control algorithm adjusts robotic arm according to current ball position. The second control algorithm calculates ball trajectory according to two sequentially captured images and positions robotic arm to intercept the ball accordingly. Next step was development of control algorithm for the second robot, which is human controlled with physical interface. In the last part we developed user interface and algorithm for real-time scoring and manipulation of the ball when it falls of the playing surface. The goal of this thesis was to create a system for real-time ball position detection, robots control, score calculation and carry out comparative analysis of both control algorithms.
|
