在基于移动边缘计算(mobile edge computing,MEC)的第5代移动通信网络(5th generation mobile communication network,5G)中,针对无线虚拟现实(virtual reality,VR)用户在小小区间频繁切换而降低业务体验质量(quality of experience,QoE...在基于移动边缘计算(mobile edge computing,MEC)的第5代移动通信网络(5th generation mobile communication network,5G)中,针对无线虚拟现实(virtual reality,VR)用户在小小区间频繁切换而降低业务体验质量(quality of experience,QoE)的问题。考虑建立多视角3维(3-dimensional,3D)视频的主动缓存、计算和通信(caching,computing and communication,3C)资源分配数学模型,并采用深度强化学习(deep reinforcement learning,DRL)算法进行求解。将多视角3D视频的主动资源分配系统建模为联合视角选择和小基站3C资源分配的马尔科夫决策过程(Markov decision process,MDP),提出了一种基于深度确定性策略梯度(deep deterministic policy gradient,DDPG)算法来寻找有效解。仿真结果表明,与另外2种算法相比,所提算法可以为无线VR用户在小小区间移动时提供更好的业务体验。展开更多
基于BEV(bird’s eye view)多传感器融合的自动驾驶感知算法近年来取得重大进展,持续促进自动驾驶的发展。在多传感器融合感知算法研究中,多视角图像向BEV视角的转换和多模态特征融合一直是BEV感知算法的重点和难点。笔者提出MSEPE-CRN(...基于BEV(bird’s eye view)多传感器融合的自动驾驶感知算法近年来取得重大进展,持续促进自动驾驶的发展。在多传感器融合感知算法研究中,多视角图像向BEV视角的转换和多模态特征融合一直是BEV感知算法的重点和难点。笔者提出MSEPE-CRN(multi-scale feature fusion and edge and point enhancement-camera radar net),一种用于3D目标检测的相机与毫米波雷达融合感知算法,利用边缘特征和点云提高深度预测的精度,实现多视角图像向BEV特征的精确转换。同时,引入多尺度可变形大核注意力机制进行模态融合,解决因不同传感器特征差异过大导致的错位。在nuScenes开源数据集上的实验结果表明,与基准网络相比,mAP提升2.17%、NDS提升1.93%、mATE提升2.58%、mAOE提升8.08%、mAVE提升2.13%,该算法可有效提高车辆对路面上运动障碍物的感知能力,具有实用价值。展开更多
作为人工智能计算机视觉领域一项重要的任务,3D人体姿态估计受到了广泛的关注,并成功地应用在人机交互、电影游戏制作等领域。然而,3D人体姿态估计仍然面临着很大的挑战,主要是人体遮挡问题和数据集视角冗余问题,这些问题严重影响了3D...作为人工智能计算机视觉领域一项重要的任务,3D人体姿态估计受到了广泛的关注,并成功地应用在人机交互、电影游戏制作等领域。然而,3D人体姿态估计仍然面临着很大的挑战,主要是人体遮挡问题和数据集视角冗余问题,这些问题严重影响了3D人体姿态估计结果精度与速度的提升。本文提出了一种基于多特征提取的3D人体姿态估计方法。首先通过采集多个相机视角下的图片数据,将所采图片数据放入2D人体关节点检测网络模型中,得到人体2D关节点。接着将采集到的人体数据输入到关节点置信度计算网络模型,得到视角图片中各个关节点的权重值。随后将2D人体关节点热图通过一个热图权重计算网络计算出热图权重,将各个视角下的权重特征计算融合得到加权后的2D人体关节点热图。最后将所得加权后的2D人体关节点热图和视角图片中各个关节点的权重值输入到三角化算法中,映射得到空间中的3D人体关节点。本文的关键思想是设计一个关节点置信度计算网络从输入图像中学习每个关节的置信度权重,同时提取了反映热图特征质量的权重矩阵,以提高遮挡视图中热图的特征质量。此外,使用感知哈希算法对Occlusion-Person数据集进行去视角实验,在保证结果准确性的同时提高了模型推理速度。本文方法是端到端可微的,可以显著地提高算法效率和鲁棒性。本文在Human3.6M和Occlusion-Person两个公共数据集上使用平均关节位置误差(Mean Per Joint Position Error,MPJPE)指标对该方法进行评估,分别取得27.3 mm和9.7 mm的结果。实验结果表明,该算法与最先进的方法相比,性能有了显著提升。展开更多
A 3D-view is helpful to instantly grasp what is presented in a drawing. There exist a variety of ways to present the same part with 3D-views. To facilitate the choice of an optimum one among them, the work divides com...A 3D-view is helpful to instantly grasp what is presented in a drawing. There exist a variety of ways to present the same part with 3D-views. To facilitate the choice of an optimum one among them, the work divides composite solid models into three categories, so as to convey the originality of design concisely and accurately by using the least ＂ engineering language＂.展开更多
Depth maps are used for synthesis virtual view in free-viewpoint television （FTV） systems. When depth maps are derived using existing depth estimation methods, the depth distortions will cause undesirable artifacts ...Depth maps are used for synthesis virtual view in free-viewpoint television （FTV） systems. When depth maps are derived using existing depth estimation methods, the depth distortions will cause undesirable artifacts in the synthesized views. To solve this problem, a 3D video quality model base depth maps （D-3DV） for virtual view synthesis and depth map coding in the FTV applications is proposed. First, the relationships between distortions in coded depth map and rendered view are derived. Then, a precisely 3DV quality model based depth characteristics is develop for the synthesized virtual views. Finally, based on D-3DV model, a multilateral filtering is applied as a pre-processed filter to reduce rendering artifacts. The experimental results evaluated by objective and subjective methods indicate that the proposed D-3DV model can reduce bit-rate of depth coding and achieve better rendering quality.展开更多
文摘在基于移动边缘计算(mobile edge computing,MEC)的第5代移动通信网络(5th generation mobile communication network,5G)中,针对无线虚拟现实(virtual reality,VR)用户在小小区间频繁切换而降低业务体验质量(quality of experience,QoE)的问题。考虑建立多视角3维(3-dimensional,3D)视频的主动缓存、计算和通信(caching,computing and communication,3C)资源分配数学模型,并采用深度强化学习(deep reinforcement learning,DRL)算法进行求解。将多视角3D视频的主动资源分配系统建模为联合视角选择和小基站3C资源分配的马尔科夫决策过程(Markov decision process,MDP),提出了一种基于深度确定性策略梯度(deep deterministic policy gradient,DDPG)算法来寻找有效解。仿真结果表明,与另外2种算法相比,所提算法可以为无线VR用户在小小区间移动时提供更好的业务体验。
文摘基于BEV(bird’s eye view)多传感器融合的自动驾驶感知算法近年来取得重大进展,持续促进自动驾驶的发展。在多传感器融合感知算法研究中,多视角图像向BEV视角的转换和多模态特征融合一直是BEV感知算法的重点和难点。笔者提出MSEPE-CRN(multi-scale feature fusion and edge and point enhancement-camera radar net),一种用于3D目标检测的相机与毫米波雷达融合感知算法,利用边缘特征和点云提高深度预测的精度,实现多视角图像向BEV特征的精确转换。同时,引入多尺度可变形大核注意力机制进行模态融合,解决因不同传感器特征差异过大导致的错位。在nuScenes开源数据集上的实验结果表明,与基准网络相比,mAP提升2.17%、NDS提升1.93%、mATE提升2.58%、mAOE提升8.08%、mAVE提升2.13%,该算法可有效提高车辆对路面上运动障碍物的感知能力,具有实用价值。
文摘作为人工智能计算机视觉领域一项重要的任务,3D人体姿态估计受到了广泛的关注,并成功地应用在人机交互、电影游戏制作等领域。然而,3D人体姿态估计仍然面临着很大的挑战,主要是人体遮挡问题和数据集视角冗余问题,这些问题严重影响了3D人体姿态估计结果精度与速度的提升。本文提出了一种基于多特征提取的3D人体姿态估计方法。首先通过采集多个相机视角下的图片数据,将所采图片数据放入2D人体关节点检测网络模型中,得到人体2D关节点。接着将采集到的人体数据输入到关节点置信度计算网络模型,得到视角图片中各个关节点的权重值。随后将2D人体关节点热图通过一个热图权重计算网络计算出热图权重,将各个视角下的权重特征计算融合得到加权后的2D人体关节点热图。最后将所得加权后的2D人体关节点热图和视角图片中各个关节点的权重值输入到三角化算法中,映射得到空间中的3D人体关节点。本文的关键思想是设计一个关节点置信度计算网络从输入图像中学习每个关节的置信度权重,同时提取了反映热图特征质量的权重矩阵,以提高遮挡视图中热图的特征质量。此外,使用感知哈希算法对Occlusion-Person数据集进行去视角实验,在保证结果准确性的同时提高了模型推理速度。本文方法是端到端可微的,可以显著地提高算法效率和鲁棒性。本文在Human3.6M和Occlusion-Person两个公共数据集上使用平均关节位置误差(Mean Per Joint Position Error,MPJPE)指标对该方法进行评估,分别取得27.3 mm和9.7 mm的结果。实验结果表明,该算法与最先进的方法相比,性能有了显著提升。
文摘A 3D-view is helpful to instantly grasp what is presented in a drawing. There exist a variety of ways to present the same part with 3D-views. To facilitate the choice of an optimum one among them, the work divides composite solid models into three categories, so as to convey the originality of design concisely and accurately by using the least ＂ engineering language＂.
基金supported by the National Natural Science Foundation of China(Grant No.60832003)Key Laboratory of Advanced Display and System Application(Shanghai University),Ministry of Education,China(Grant No.P200902)the Key Project of Science and Technology Commission of Shanghai Municipality(Grant No.10510500500)
文摘Depth maps are used for synthesis virtual view in free-viewpoint television （FTV） systems. When depth maps are derived using existing depth estimation methods, the depth distortions will cause undesirable artifacts in the synthesized views. To solve this problem, a 3D video quality model base depth maps （D-3DV） for virtual view synthesis and depth map coding in the FTV applications is proposed. First, the relationships between distortions in coded depth map and rendered view are derived. Then, a precisely 3DV quality model based depth characteristics is develop for the synthesized virtual views. Finally, based on D-3DV model, a multilateral filtering is applied as a pre-processed filter to reduce rendering artifacts. The experimental results evaluated by objective and subjective methods indicate that the proposed D-3DV model can reduce bit-rate of depth coding and achieve better rendering quality.
