Vision dataset for ColRobot WP6 use-case

<p>A specific 2D sensing system, comprising of a 2D camera, a backlit feeder and its associated processing has been developed for the ColRobot kitting use-case. This sensing system is integrated with the gripper and the robot.</p> <p>The objective of the sensing system is to detect and locate the kitting parts (notably screws, washers and nuts) in order to grasp them.</p> <p>The sensing system must be able:</p> <ul> <li>To be rapidly reconfigured for parts of new dimensions (in case of evolution of the kitting specifications), without reprogramming.</li> <li>To be able to isolate parts among similar (but not identical) parts.</li> <li>To be able to avoid detecting parts that are overlapping.</li> </ul> <p>To be able to avoid detecting parts that are too close to each other (clearance should be above 6mm).</p> <p>Constraints are imposed on the processing algorithm to avoid grasping issues:</p> <ul> <li>Do not detect overlapping parts.</li> <li>Do not detect parts too close to another part (identical or not), i.e. keep 6mm clearance around any detected part.</li> </ul> <p>All input pictures are taken with the same sensor, at the same distance to the backlit feeder.</p> <p>Processing parameters are the same for all images.</p> <p>The list of parts to be detected with their parameters is given below in the Data input section.</p> <p>For all input images, all parts given in the input parameters are searched for and an output image is created if found highlighting a point of interest of the part, and an associated oriented frame.</p> <p>On some input images, additional (non-referenced) parts are present in order to test the discrimination properties of the algorithm (false positives).</p> <p>The dataset takes the following format:</p> <ul> <li>Input data: <ul> <li>List of input images.</li> <li>Table of parameters to be given to the processing algorithm for each input image. <ul> <li>For each part, a part type, and an ordered list of dimensions (semantics depending on the part type).</li> </ul> </li> </ul> </li> <li>Output data: <ul> <li>A structured folder of output images <ul> <li>For each input images, a clone of the input image is created for each part described in the input parameters, only if found, and highlighting: <ul> <li>A point of interest (static in the part frame for a given part type).</li> <li>An associated frame describing the part orientation.</li> </ul> </li> </ul> </li> </ul> </li> </ul>