A decentralized task planning algorithm is proposed for heterogeneous unmanned aerial vehicle(UAV)swarm with different capabilities.The algorithm extends the consensus-based bundle algorithm(CBBA)to account for a more...A decentralized task planning algorithm is proposed for heterogeneous unmanned aerial vehicle(UAV)swarm with different capabilities.The algorithm extends the consensus-based bundle algorithm(CBBA)to account for a more realistic and complex environment.The extension of the algorithm includes handling multi-agent task that requires multiple UAVs collaboratively completed in coordination,and consideration of avoiding obstacles in task scenarios.We propose a new consensus algorithm to solve the multi-agent task allocation problem and use the Dubins algorithm to design feasible paths for UAVs to avoid obstacles and consider motion constraints.Experimental results show that the CBBA extension algorithm can converge to a conflict-free and feasible solution for multi-agent task planning problems.展开更多
针对不同型号车辆外观差异较小,车辆检索困难的问题,构建一种两阶段细粒度车辆检索算法。该算法选择包含有效信息的特征,第一阶段通过广义平均池化(Generalized Mean Pooling)产生全局特征描述子,最后通过欧氏距离法得到初次检索结果。...针对不同型号车辆外观差异较小,车辆检索困难的问题,构建一种两阶段细粒度车辆检索算法。该算法选择包含有效信息的特征,第一阶段通过广义平均池化(Generalized Mean Pooling)产生全局特征描述子,最后通过欧氏距离法得到初次检索结果。第二阶段通过Faster R-CNN预测目标区域的类别得分和位置坐标,在初次检索结果中找到与该查询类别相同的目标区域,并结合扩展查询(Query Expansion)对目标区域特征再次进行欧氏距离计算,检索出最终相似的图像。实验结果证明,该方法在细粒度车型数据集上取得了较好的效果。展开更多
文摘A decentralized task planning algorithm is proposed for heterogeneous unmanned aerial vehicle(UAV)swarm with different capabilities.The algorithm extends the consensus-based bundle algorithm(CBBA)to account for a more realistic and complex environment.The extension of the algorithm includes handling multi-agent task that requires multiple UAVs collaboratively completed in coordination,and consideration of avoiding obstacles in task scenarios.We propose a new consensus algorithm to solve the multi-agent task allocation problem and use the Dubins algorithm to design feasible paths for UAVs to avoid obstacles and consider motion constraints.Experimental results show that the CBBA extension algorithm can converge to a conflict-free and feasible solution for multi-agent task planning problems.
文摘针对不同型号车辆外观差异较小,车辆检索困难的问题,构建一种两阶段细粒度车辆检索算法。该算法选择包含有效信息的特征,第一阶段通过广义平均池化(Generalized Mean Pooling)产生全局特征描述子,最后通过欧氏距离法得到初次检索结果。第二阶段通过Faster R-CNN预测目标区域的类别得分和位置坐标,在初次检索结果中找到与该查询类别相同的目标区域,并结合扩展查询(Query Expansion)对目标区域特征再次进行欧氏距离计算,检索出最终相似的图像。实验结果证明,该方法在细粒度车型数据集上取得了较好的效果。
