This research implements robotic vision image recognition techniques of comparing actual pictures to a theoretical model. Specifically, such a comparison is used for pose estimation of a baseball. Because the seams of the baseball follow a predictable curve, the model can use the seams to recognize orientation from a twodimensional image. This simplification of input data allows for greater computational efficiency in computer vision. Existing research is adapted to create a model of the baseball seams for this project. An apparatus from previous research capable of rotating a baseball along two mutually perpendicular axes of rotation is modified and used to obtain a data set of pictures of the baseball in different orientations. Using MATLAB, the pictures are analyzed to extract critical points along the seams. The model is rotated according to the rotation of the ball in the picture, and the sum of the squared residuals per point is calculated as a measure of the usefulness of the model. This procedure is implemented in an autonomic system capable of determining the relative pose of a baseball in a picture under any possible rotation. Then the entire system is tested by comparing calculated rotational components to actual rotational components, and the system is able to solve for the relative pose of the baseball within 5.39 degrees of rotational error. In practical applications this amount of error is acceptable for manipulating objects, especially due to the unique efficiency of this system. 

Get Inspired Check out our sample "Project Summary" page and get tips for using Google Sites in our resource guide. 