Robust reflective surface inspection approaches

We have developed several approaches for fast anomaly detection on reflective surfaces.

Application to car damage inspection:

A software solution was designed for a private company. The system is composed of a rig equipped with a pattern projector, five cameras and distance sensors. As the car drives through the rig, the system detects the car and starts capturing synchronized video and distance streams. Advanced computer vision algorithms are then used to process the video streams, automatically detect car parts and the car damage on each part.

Damages on the car surface cause deformation of the projected pattern. Computer vision methods are used to detect certain frequency changes in the projected pattern which belong to the damaged regions. The system first quickly detects potential damaged regions, which are then verified by advanced machine learning and image processing techniques. The extent of deformed pattern around damage center is calculated in pixels and converted by a multi-camera system into millimeters and then into a damage size category. Motion consistency and multi-frame verification is applied to keep track of each detected dent identity through time and ensure a single dent is counted only once.


Here's a beta 2016 version: