Automatic bin picking uses camera to detect where bins are and then manouvers picking. In this work, an automatic ellipse extractor is developed which detects bins as ellipses in the image. Ellipse parameter is first discretized which fits available parallel HW. A prior mask is then generated by matching edge statistics over full image. The algorithm fits ellipses in two steps: first using low-resolution images and then at high resolution using constrained parameter domain. Fit is found by rapid ellipse contour rasterization step which is executed in parallel for candidates across the image. It is very fast as it is derived using Bresenham principles. Edge responses are then be evaluated for each attempt and best peaks in cost function produce an ellipse. Finally the overlap tests are used to filter out impossible ellipses and results are visualized. Upside of the algorithm is scalability for modern parallel HW and avoiding feature extraction steps completely.
Algorithm output example for input image.
Algorithm output example for input image.
Algorithm output example for input image.
Algorithm output example for input image.