Challenge

In a manufacturing context the main requirement of the task is to ensure that a set of cables is correctly positioned in their corresponding slots separated by boundary lines. Each cable has a fixed position and is not allowed to cross the boundaries between slots due to security constrains.

Solution

We cast the problem as the task of semantic segmentation and trained multiple neural networks each designed to provide specific information. The combined solution automatically identifies individual cables, verifies if they lie in the right slots and also checks if there are boundary crossings. The pictures on the right illustrate one possible scenario including 3 different cables. From left to right, top down, the first image presents an input to the system. The second and third image show computed segmentation mask and masked input image. Finally, the last image shows extracted information necessary to validate the state of cables.

Impact

The implemented solution is both fast and accurate for verification of cable positions fitting all technical project requirements.