Fault-tolerant and error-detection designs for a computer vision system

Computer vision is one of the modern inspection approaches employed in automated manufacturing systems. The measurement accuracy of a computer vision-system is highly correlated with variational and unstable factors in the environment. By utilizing the redundancy concept. this paper presents multi-camera designs for a computer vision system to reduce measurement errors. The design schemes discussed in this paper are the single-camera design. fault-tolerant design (two cameras) and error-detectable design (three cameras). Performance of the design schemes is evaluated via experiments and statistical analysis. Based on the neural network concept a prototype module is implemented to analyze the experimental data and error distribution of each design scheme. In addition, after integrating the learning capability into the module. feasibility and effectiveness of the proposed designs are validated. It is (he pal of this paper to propose the design schemes as well as an application module for computer vision systems to promote measurement accuracy and information reliability.