摘要
作为典型的NP问题,双目立体视觉中的匹配一直是研究的热点。提出一种适于求解立体匹配问题的并行、迭代、基于信息反馈的多群体蚁群算法。该算法中子群体完成每个扫描行对应的子优化问题;协调者收集并分析所有子群体的结果,通过投票决定可能匹配的边缘,进而把这种信息转换成信息素传播给子群体,如此迭代直到优化结束。还讨论了双目立体匹配过程中的约束问题,并给出算法的收敛性证明和仿真实验。
As a typical NP-hard problem,stereo matching is and will be a hot topic to research.A parallel,iterative and feedback algorithm for stereo matching was proposed based on multiple ant colony optimization.In this algorithm,the matching of every scan-line was solved by single ant colony and the master collected and analyzed the results from colonies and got matched edges by voting,then reconstructed and broadcast the pheromone to every colony.The colony started the new iteration on the new pheromone from the master till a stopping criterion was met.The constraints that the optimization must be obeyed during the stereo matching procedure were discussed.The convergence proof and experiments were given.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2012年第5期999-1004,1015,共7页
Journal of System Simulation
关键词
双目立体匹配
多群体蚁群算法
迭代
置信度传播
binocular stereo matching
multiple ant colony optimization
iteration
reliability propagation
