Switch policy is essential for small cells to properly serve variable number of users in an energy efficient way.However,frequently switching small cell base stations(SBSs) may increase the network operating cost,espe...Switch policy is essential for small cells to properly serve variable number of users in an energy efficient way.However,frequently switching small cell base stations(SBSs) may increase the network operating cost,especially when there is an nonnegligible start-up energy cost.To this end,by observing the variety of user number,we focus on the design of a switch policy which minimize the cumulative energy consumption.A given user transmission rate is guaranteed and the capability of SBSs are limited as well.According to the knowledge on user number variety,we classify the energy consumption problem into two cases.In complete information case,to minimize the cumulative energy consumption,an offline solution is proposed according to critical segments.A heuristic algorithm for incomplete information case(HAIIC) is proposed by tracking the difference of cumulative energy consumption.The upper bound of the Energy Consumption Ratio(ECR) for HAIIC is derived as well.In addition,a practical Q-learning based probabilistic policy is proposed.Simulation results show that the proposed HAIIC algorithm is able to save energy efficiently.展开更多
针对人体关节点序列的连续行为识别问题,提出了一种基于BSCPs-RF(B-spline control points-random forest)的人体关节点信息行为识别与预测方法.首先采用局部线性回归与单帧关节点归一化法预处理关节点序列,以此消除抖动噪声、位移和尺...针对人体关节点序列的连续行为识别问题,提出了一种基于BSCPs-RF(B-spline control points-random forest)的人体关节点信息行为识别与预测方法.首先采用局部线性回归与单帧关节点归一化法预处理关节点序列,以此消除抖动噪声、位移和尺度的影响;然后以B样条曲线控制点作为速度无关的关节点序列特征,并采用同步语音提示词法标注实时行为序列以提高样本采集效率;最后采用基于随机森林的行为识别与预测方法,并以集成学习方法优化多分类器组合以提高识别精度.实验分析了不同参数值对识别效果的影响,并分别在测试数据库MSR-Action3D以及RGB-D设备采集的实时数据集中进行测试.结果显示,MSR-Action3D测试结果优于部分先前方法,而实时数据测试中该方法具有很高的识别精度,进而验证了该方法的有效性.展开更多
基金partially supported by National Key Project of China under Grants No. 2013ZX03001007-004National Natural Science Foundation of China under Grants No. 61102052,61325012,61271219,91438115 and 61221001
文摘Switch policy is essential for small cells to properly serve variable number of users in an energy efficient way.However,frequently switching small cell base stations(SBSs) may increase the network operating cost,especially when there is an nonnegligible start-up energy cost.To this end,by observing the variety of user number,we focus on the design of a switch policy which minimize the cumulative energy consumption.A given user transmission rate is guaranteed and the capability of SBSs are limited as well.According to the knowledge on user number variety,we classify the energy consumption problem into two cases.In complete information case,to minimize the cumulative energy consumption,an offline solution is proposed according to critical segments.A heuristic algorithm for incomplete information case(HAIIC) is proposed by tracking the difference of cumulative energy consumption.The upper bound of the Energy Consumption Ratio(ECR) for HAIIC is derived as well.In addition,a practical Q-learning based probabilistic policy is proposed.Simulation results show that the proposed HAIIC algorithm is able to save energy efficiently.
文摘针对人体关节点序列的连续行为识别问题,提出了一种基于BSCPs-RF(B-spline control points-random forest)的人体关节点信息行为识别与预测方法.首先采用局部线性回归与单帧关节点归一化法预处理关节点序列,以此消除抖动噪声、位移和尺度的影响;然后以B样条曲线控制点作为速度无关的关节点序列特征,并采用同步语音提示词法标注实时行为序列以提高样本采集效率;最后采用基于随机森林的行为识别与预测方法,并以集成学习方法优化多分类器组合以提高识别精度.实验分析了不同参数值对识别效果的影响,并分别在测试数据库MSR-Action3D以及RGB-D设备采集的实时数据集中进行测试.结果显示,MSR-Action3D测试结果优于部分先前方法,而实时数据测试中该方法具有很高的识别精度,进而验证了该方法的有效性.