Industry project · UNIEYE · Feb–Jul 2020

Pattern inspection with two industrial cameras

An inspection program on a Jetson Nano. It found circular marks in images from two Basler cameras, measured where they sat, and passed or failed each part.

Role: User interface and communication
Team: 2 students
Hardware: Jetson Nano, 2 × Basler acA2500-14gm
Stack: C++, Qt

Inspection program window with two camera views of circular marks, pass results, counters and offset plots — The inspection program during a run. Both cameras matched the target mark, with scores of 97.48 and 96.55, and the part passed. The right panel counts failures by type; the plots track the mark's offset over successive parts.

How it worked

1 Capture Frames were downscaled from 2592×1944 to 324×243. At 648×486 the Jetson Nano could not keep up.
2 Filter A Gaussian filter and Canny edge detection, both implemented by hand.
3 Find circles Contours with a radius of 10–150 px, at least 15 px apart, whose distance from the center varied by less than 15%.
4 Judge The mark's score and X/Y position were checked against a zero-set reference, an allowed pixel offset and an accuracy threshold.

My part

On a team of two, my part was the Qt interface shown above and the communication layer.

The communication layer linked the Jetson Nano to a host PC. The system was built as a prototype.