Spatial-temporal visual inspection

Background

Visual inspections are time-consuming and can be influenced by human bias and error. Many current anomaly-detection methods compare a structure’s present condition to a presumed “normal” state, but because every piece of civil infrastructure is complex and unique, defining the “normal” is difficult.

Computer vision can detect and segment defects like cracks, spalling, and corrosion using predefined categories. However, this semantic approach requires those categories to be known in advance. In structural health monitoring, that is often unrealistic: failures are rare, hard to model, and can occur in unexpected ways. As a result, rare failures may be identified poorly, and unexpected failures may be missed altogether.

Description of the invention

Researchers at the University of Waterloo have invented a novel computer vision system containing two Novel View Synthesis (NVS) models. Each of the NVS models is constructed from image data representing current and previous inspection of a structure. The models are then synchronized such that for every camera and camera pose, a pair of pixel-aligned images of the same structure can be built, differing only in when the inspection took place. This enables pixel-wise alignment of the images from different time windows to detect any changes/anomalies. The invention allows for rapid assessment of changes, and through performing repeated inspections, the rate of change, which is a critical dimension for evaluating structural safety in a structure, can be accurately measured.

Advantages

Rapid analysis of images taken from a structure enables onsite and real-time inspection, which reduces the cost (by eliminating the need for revisiting the site for re-inspection and filling the data gap) and increases safety (potential issues are detected more accurately and much faster). The methodology also enables remote piloting of drones, which further reduces the cost of inspection and increases safety.

Potential applications

• Structural health monitoring
• Real time inspection of civil structures
• Telecommunication tower and power pylon inspections (the invention enables detection of bolt-loosening)
• Drone-based visual inspection