INITIAL-VALUE SEARCH METHOD FOR MULTIPLE VIEWPOINT STEREO MEASUREMENTUSING GENETIC ALGORITHMS

There has already been a report on the method to improve stereo measur ement accuracy by integrating stereo data from multiple viewpoints, wh ich employs a local optimization technique, the nonlinear optimization method using iterative calculation. The problem is that estimated res ults depend on the initial values of the iterative calculation. If the calculation begins with unsuitable values, a correct estimation canno t be made. This paper presents a new method utilizing a genetic algori thm to search for initial values for the nonlinear optimization proces s in multiple viewpoint stereo measurement. In this method, a one-dime nsional matrix is created. Called an ''individual,'' it consists of pa rameters to be determined. Next, a population is formed which consists of several individuals. Then three kinds of genetic operations are pe rformed on this population, crossover, mutation and selection, which e volve the population to fit the environment. This evolution process is repeated and the evolved result is used as the initial values for the iteration calculation. The efficiency of this method is confirmed by simulation and an outdoor experiment.