Automated Knowledge-Based System for Stereo Video Metrology

Abstract

A knowledge-based system has been developed to help inexperienced users make measurements from stereo video images. The purpose of the system is to automate much of the routine functions and decision making in photogrammetric measurements on a personal computer (PC). The system can perform the following functions: (1) Check the validity of the input data; (2) warn of weak geometric conditions; (3) provide guidance, diagnostics, and counseling during success and failure modes; (4) conduct robust blunder detection; and (5) perform accuracy analysis through error propagation. The result was the development of a user-friendly vision system that can be used productively without in-depth knowledge of photogrammetry. Experimental results showed that the PC-based vision system achieved a potential accuracy of about one pixel on the image plane for planar coordinates. Lower measurement accuracy in the range of 4–5 pixels was obtained for the depth direction because of the intersection geometry and accuracy limitations in manual image matching. The statistical analysis scheme, based on random error propagation of the image coordinates, was a realistic accuracy estimator. Calculated three-dimensional (3D) measurement errors consistently fell within three times the estimated standard errors (3σ). Comparison with actual survey measurements showed that distances could be measured with an accuracy of better than 2 pixels, while volume and surface area were measured to within 3. Image scale, base/object distance ratio, number and distribution of control points, and accuracy limitation in manual matching had a significant impact on the measurement accuracy.