Machine vision algorithms for line-scan TDI cameras.

摘要

The goal of industrial machine vision systems has been to replace human vision capabilities in the area of automated inspection of manufactured parts and processes. Advantages of machine vision systems over human inspection include resolution, speed, reliability, consistency and long-term cost. In this thesis we address issues associated with the automated inspection of web manufacturing processes, in which several camera are used to inspect material that is produced in roll form, such as paper products, textiles, insulating film, etc. Our work concentrates on the use of cameras with line-scan CCD sensors, and which use the Time-Delay and Integration technique to produce high contrast images at high speed and in ordinary lighting conditions. In particular, we look at in-camera preprocessing of the high bandwidth sensor data with the objective of limiting the output bandwidth of video data from the camera.; We contribute to two related preprocessing issues, The first is a novel technique to replace the shaft encoders, used to provide the image velocity to the TDI circuitry inside the camera, by easily implemented algorithms that use the direct output of the TDI sensor for an indirect measurement of image velocity. In this work, we develop several algorithms that can adjust the TDI charge transfer speed by measuring the frequency properties of the image output of the camera. We prove the effectiveness of our algorithms by implementing them in a commercial TDI camera.; For the second part of our work we develop low complexity hash functions to compare repeating patterns with themselves, and show how these can be used to detect defects on patterned images. We prove the efficiency of these algorithms through simulations and implementations on a camera. We also develop a streaming version of the sum and difference matrices approach, a method that has proven to be very effective to locate defects on fabrics. Simulations demonstrate the detection of defects on many types of fabrics using algorithms that are efficient to implement in commercial TDI cameras.