In digital video analysis, browse, retrieval and query, shot is incapable of meeting needs. Scene is a cluster of a series of shots, which partially meets above demands. In this paper, an algorithm of video scenes clu...In digital video analysis, browse, retrieval and query, shot is incapable of meeting needs. Scene is a cluster of a series of shots, which partially meets above demands. In this paper, an algorithm of video scenes clustering based on shot key frame sets is proposed. We use X^2 histogram match and twin histogram comparison for shot detection. A method is presented for key frame set extraction based on distance of non adjacent frames, further more, the minimum distance of key frame sets as distance of shots is computed, eventually scenes are clustered according to the distance of shots. Experiments of this algorithm show satisfactory performance in cor rectness and computing speed.展开更多
街道场景视频实例分割是无人驾驶技术研究中的关键问题之一,可为车辆在街道场景下的环境感知和路径规划提供决策依据.针对现有方法存在多纵横比锚框应用单一感受野采样导致边缘特征提取不充分以及高层特征金字塔空间细节位置信息匮乏的...街道场景视频实例分割是无人驾驶技术研究中的关键问题之一,可为车辆在街道场景下的环境感知和路径规划提供决策依据.针对现有方法存在多纵横比锚框应用单一感受野采样导致边缘特征提取不充分以及高层特征金字塔空间细节位置信息匮乏的问题,本文提出锚框校准和空间位置信息补偿视频实例分割(Anchor frame calibration and Spatial position information compensation for Video Instance Segmentation,AS-VIS)网络.首先,在预测头3个分支中添加锚框校准模块实现同锚框纵横比匹配的多类型感受野采样,解决目标边缘提取不充分问题.其次,设计多感受野下采样模块将各种感受野采样后的特征融合,解决下采样信息缺失问题.最后,应用多感受野下采样模块将特征金字塔低层目标区域激活特征映射嵌入到高层中实现空间位置信息补偿,解决高层特征空间细节位置信息匮乏问题.在Youtube-VIS标准库中提取街道场景视频数据集,其中包括训练集329个视频和验证集53个视频.实验结果与YolactEdge检测和分割精度指标定量对比表明,锚框校准平均精度分别提升8.63%和5.09%,空间位置信息补偿特征金字塔平均精度分别提升7.76%和4.75%,AS-VIS总体平均精度分别提升9.26%和6.46%.本文方法实现了街道场景视频序列实例级同步检测、跟踪与分割,为无人驾驶车辆环境感知提供有效的理论依据.展开更多
基金Supported by the Natural Science Foundation ofHubei Province(2004ABA174)
文摘In digital video analysis, browse, retrieval and query, shot is incapable of meeting needs. Scene is a cluster of a series of shots, which partially meets above demands. In this paper, an algorithm of video scenes clustering based on shot key frame sets is proposed. We use X^2 histogram match and twin histogram comparison for shot detection. A method is presented for key frame set extraction based on distance of non adjacent frames, further more, the minimum distance of key frame sets as distance of shots is computed, eventually scenes are clustered according to the distance of shots. Experiments of this algorithm show satisfactory performance in cor rectness and computing speed.
文摘街道场景视频实例分割是无人驾驶技术研究中的关键问题之一,可为车辆在街道场景下的环境感知和路径规划提供决策依据.针对现有方法存在多纵横比锚框应用单一感受野采样导致边缘特征提取不充分以及高层特征金字塔空间细节位置信息匮乏的问题,本文提出锚框校准和空间位置信息补偿视频实例分割(Anchor frame calibration and Spatial position information compensation for Video Instance Segmentation,AS-VIS)网络.首先,在预测头3个分支中添加锚框校准模块实现同锚框纵横比匹配的多类型感受野采样,解决目标边缘提取不充分问题.其次,设计多感受野下采样模块将各种感受野采样后的特征融合,解决下采样信息缺失问题.最后,应用多感受野下采样模块将特征金字塔低层目标区域激活特征映射嵌入到高层中实现空间位置信息补偿,解决高层特征空间细节位置信息匮乏问题.在Youtube-VIS标准库中提取街道场景视频数据集,其中包括训练集329个视频和验证集53个视频.实验结果与YolactEdge检测和分割精度指标定量对比表明,锚框校准平均精度分别提升8.63%和5.09%,空间位置信息补偿特征金字塔平均精度分别提升7.76%和4.75%,AS-VIS总体平均精度分别提升9.26%和6.46%.本文方法实现了街道场景视频序列实例级同步检测、跟踪与分割,为无人驾驶车辆环境感知提供有效的理论依据.
