A computer vision system that locates and identifies grading defects in rough hardwood lumber in a species-independent manner is described in detail. It consists of a low-level module that performs segmentation and extracts region properties, and a high-level module that identifies the type of defect present in each of the regions passed from the low-level module and extracts the appropriate characteristics associated with each defect. The system has been designed using a knowledge-based approach using a blackboard framework and has been tested on a number of boards from four hardwood species. The current system can detect four of the most common types of defects: knots, holes, wane, and splits/checks. Although it has limited recognition capabilities, the results suggest that species-independent methods can be found for accomplishing the required tasks.
展开▼
