Jocc's profile - activity

Yes. But there will be more similar workpieces, and i have to distinguish the workpieces (e.g. on the picture above there are two different resistors with marks on their top). To clarify, the main purpose of the system: on an assembly line the parts are coming on trays to the operator and the system have to check that every parts are on the correct place and every part is the correct (prescribed) part.

The small rectangle-like parts (resistors) has too few good features to track, they need template matching, but tempmatch is slow(er than feature matching). The workpieces can be rotated around all axes of the space. Orientation around the "camera axis" can be handled by the algorithm. My idea is to use many reference images, photos of the parts around their own axis in 10-20 degree steps.

Hi! I have to detect parts in a bulk storage at high speed (at least 2-3 fps). I need high resolution because the working space is rather large (approx. 0.5x0.8 m) and the parts are small (centimeters range), so my camera resolution is 5MPix. I've tested the feature-based detection algorithms, SURF, SIFT and ORB were the best (all of them was stable, ORB is fast enopugh), but some of my workpieces are very simple, has not too much features to detect. I've tested the template matching (tempMatch), but it is very slow in high resolution and because of the bulk storage the algorithm must deal with changes in scale, rotation and aspect ratio. Has anybody experience with this type of task ? Which is the best algorithm ? Thanks in advance.