基于视觉注意机制及视嗅觉融合的气体泄漏源自主搜寻研究

项目名称： 基于视觉注意机制及视嗅觉融合的气体泄漏源自主搜寻研究

项目编号： No.60802051

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 数理科学和化学

项目作者： 曾明

作者单位： 天津大学

项目金额： 20万元

中文摘要： 围绕气味源定位问题，本课题组依据不同的搜寻线索系统地开展了三个方面的主题研究，即"基于视觉信息的气味源定位研究"、"基于羽流信息的气味源定位研究"、"基于多感知信息融合的气味源定位研究"。其中代表性的工作如下：综述了当前国内外气味源定位方面的研究成果；借鉴现有烟羽模型的部分思想并结合计算流体力学相关理论，建立了用于验证气味源定位算法的室内二维湍流烟羽模型；为了解决视觉信息处理中的"瓶颈问题"，提出了用于气味源搜寻的任务驱动视觉注意机制计算模型；提出了一种优化的特征调制策略，使得待搜寻目标更容易凸显出来；提出了气味包路径估计方法，该方法可用于指导机器人进行气味源的搜寻和最终气味源确认；为了更有效地利用风场和浓度场信息同时避免陷入局部极值，提出了两种烟羽跟踪算法，即改进的蚁群优化结合逆风算法和基于概率粒子群优化算法；分析了多机器人气味源定位中的探索和利用特性，提出了二者平衡的指标和分类方法；提出了基于包容体系结构的融合视/嗅觉信息的气味源定位方法；提出了基于启发式思想结合气味质量通量计算的气味源确认算法。大量仿真或真实机器人实验验证了上述方法的有效性和先进性。

中文关键词： 气味源定位；视觉注意机制；烟羽跟踪；多源信息融合

英文摘要： Centering on the problem of odor source localization (OSL), systematic studies have been performed according to three different clues, such as visual information, plume information and multi-source information. The main achievements can be concluded as follows: The state-of-the-art methods for OSL were surveyed. A two-dimension indoor turbulent plume model used for verifying the OSL algorithm was built by combining the theory of computational fluid dynamics and partial ideas of existing plume models. In ordor to solve the " bottleneck problem " of visual information processing, a top-down visual attention model for OSL was proposed. We derived an optimal feature modulation strategy to maximize the relative salience of the target. A method for estimating the historical trajectory of the odor patch was put forward, which can be applied to search for the odor source and finally to declare the odor source. In ordor to efficiently use the wind and odor information and avoid falling into local extrema, Upwind surge combined with Modified Ant Colony Optimization (U-MACO) as well as Probability Particle Swarm Optimization (P-PSO), were proposed. The characteristics of exploration and exploitation (E-E) in multi-robot based odor source localization were analyzed. The index and classification method for balancing the E-E were proposed. A fusion method of vision and olfaction based on subsumption architecture was presented to accomplish the OSL task in the inturbulence dominated airflow environments.A gas source declaration algorithm was brought forward by combining heuristic thoughts and odor mass throughput calculation. The feasibility and advanced nature of these methods have been verified in a large numberof simulation or real robot experiments.

英文关键词： odor source localization; visual attention mechanisms;plume tracking; multi-source information fusion