The inspection software for the VideoAOI components in the Flying Probe Testers is being developed here at the chair of applied computer science IV (Lehrstuhl für Praktische Informatik IV) under the guidance of professor Effelsberg. VideoAOI is short hand for Video Automatic Optical Inspection -- as opposed to traditional Automatic Optical Inspection based on still images. The optical inspection shall precede the inspection by the Flying Probe Tester. While the PCB is conveyed into the tester an installed camera captures video recordings of the board. These can be analyzed by the software in realtime. In doing so, the software employs algorithms for image segmentation, pattern recognition and texture analysis, previously developed and implemented at our chair.

As an initial step the exact position and orientation of the PCB on the conveyor belt has to be determined. For this purpose PCBs provide designated points of reference called fiducial markers, designed in a way so that they're efficiently detectable by means of computer vision. The fiducial markers on a board form a coordinate system that allows to describe the position of placed components like resistors and capacitors. The positions of the markers are also relevant for the Flying Probe Testers as they're required to accurately determine points for connecting the sensors.

In the next step the recordings shall be compared to the saved description of the board to detect misplaced, incorrect, or completely missing components. The advantage of analyzing video recordings over the analysis of still images lies in the fact that the parts are being recorded while moving along the conveyor, which allows them to be viewed from different perspectives. This increase of available information enables a higher probability of detecting errors. In addition to that, they allow for the application of "structure from motion" approaches or the reconstruction of depth.

Errors detected by the software will subsequently be transmitted to the Flying Probe Tester. It can then test the erroneous parts more thoroughly or possibly skip the testing of correct parts. This is expected to result in a reduction of total time spent for functional testing and in increased confidence in the output of the tests.