Deflectometry-based detection of specific reflective surface anomalies

Lojze Žust (2018) Deflectometry-based detection of specific reflective surface anomalies. Prešeren awards for students.

Preview

PDF Download (6Mb)

Abstract

In this work we propose a new deflectometry-based anomaly detection approach applicable to reflective surfaces. Classic deflectometry methods detect surface anomalies by performing partial 3D surface reconstruction and differencing it with a pre-recorded reference model of the observed object. Most of these methods require accurate calibration between the pattern projector, camera and the inspected object. In contrast, our anomaly detection approach is capable of fast detection without the need for a 3D reconstruction. Since the proposed method can be trained on annotated anomaly examples, reference objects and accurate calibration are not required. Furthermore, the developed method brings forward an entirely new approach to anomaly detection and is made up of two parts. The first part is based on semantic segmentation and performs pixel-wise anomaly classification. We utilize the power of deep models for this purpose. The second part is a robust mechanism for anomaly localization from the segmentation mask, capable of extracting even largely overlapping detections. Preliminary analysis and experimental evaluation were performed to justify the architecture and hyperparameters of our deep semantic segmentation model. The final model was trained and evaluated on the problem of dent detection in car roofs, where a significant improvement over the state of the art commercial method has been shown. Our model is 17x faster and almost 50 % more accurate than the commercial method and achieves a precision, recall and F-score of 88 %.

Actions (login required)