在复杂道路场景下,车辆目标之间频繁遮挡、车辆目标之间相似的外观、目标整个运动过程中采用静态预设参数都会引起跟踪准确率下降等问题。该文提出了一种基于车辆外观特征和帧间光流的目标跟踪算法。首先,通过YOLOv5算法中的YOLOv5x网...在复杂道路场景下,车辆目标之间频繁遮挡、车辆目标之间相似的外观、目标整个运动过程中采用静态预设参数都会引起跟踪准确率下降等问题。该文提出了一种基于车辆外观特征和帧间光流的目标跟踪算法。首先,通过YOLOv5算法中的YOLOv5x网络模型获得车辆目标框的位置信息;其次,利用RAFT (recurrent all-pairs field transforms for optical flow)算法计算当前帧和前一帧之间的光流,并根据得到的位置信息对光流图进行裁剪;最后,在卡尔曼滤波过程中利用帧间光流进行补偿得到更精确的运动状态信息,并利用车辆外观特征和交并比特征完成轨迹匹配。实验结果表明,基于车辆外观特征和帧间光流的目标跟踪算法在MOT16数据集上表现良好,相较于跟踪算法DeepSORT,成功跟踪帧数占比提高了1.6%,跟踪准确度提升了1.3%,跟踪精度提升了0.6%,改进的车辆外观特征提取模型准确率在训练集和验证集上分别提高了1.7%、6.3%。因此,基于高精度的车辆外观特征模型结合关联帧间光流的运动状态信息能够有效实现交通场景下的车辆目标跟踪。展开更多
Based on the Independent Continuous Mapping method （ICM）, a topological optimization model with continuous topological variables is built by introducing three filter functions for element weight, element allowable s...Based on the Independent Continuous Mapping method （ICM）, a topological optimization model with continuous topological variables is built by introducing three filter functions for element weight, element allowable stress and element stiffness, which transform the 0-1 type discrete topological variables into continuous topological variables between 0 and 1. Two methods for the filter functions are adopted to avoid the structural singularity and recover falsely deleted elements： the weak material element method and the tiny section element method. Three criteria （no structural singularity, no violated constraints and no change of structural weight） are introduced to judge iteration convergence. These criteria allow finding an appropriate threshold by adjusting a discount factor in the iteration procedure. To improve the efficiency, the original optimization model is transformed into a dual problem according to the dual theory and solved in its dual space. By using MSC/Nastran as the structural solver and MSC/Patran as the developing platform, a topological optimization software of frame structures is accomplished. Numerical examples show that the ICM method is very efficient for the topological optimization of frame structures.展开更多
文摘在复杂道路场景下,车辆目标之间频繁遮挡、车辆目标之间相似的外观、目标整个运动过程中采用静态预设参数都会引起跟踪准确率下降等问题。该文提出了一种基于车辆外观特征和帧间光流的目标跟踪算法。首先,通过YOLOv5算法中的YOLOv5x网络模型获得车辆目标框的位置信息;其次,利用RAFT (recurrent all-pairs field transforms for optical flow)算法计算当前帧和前一帧之间的光流,并根据得到的位置信息对光流图进行裁剪;最后,在卡尔曼滤波过程中利用帧间光流进行补偿得到更精确的运动状态信息,并利用车辆外观特征和交并比特征完成轨迹匹配。实验结果表明,基于车辆外观特征和帧间光流的目标跟踪算法在MOT16数据集上表现良好,相较于跟踪算法DeepSORT,成功跟踪帧数占比提高了1.6%,跟踪准确度提升了1.3%,跟踪精度提升了0.6%,改进的车辆外观特征提取模型准确率在训练集和验证集上分别提高了1.7%、6.3%。因此,基于高精度的车辆外观特征模型结合关联帧间光流的运动状态信息能够有效实现交通场景下的车辆目标跟踪。
基金The project supported by the National Natural Science Foundation of China (10472003)Beijing Natural Science Foundation (3042002)
文摘Based on the Independent Continuous Mapping method （ICM）, a topological optimization model with continuous topological variables is built by introducing three filter functions for element weight, element allowable stress and element stiffness, which transform the 0-1 type discrete topological variables into continuous topological variables between 0 and 1. Two methods for the filter functions are adopted to avoid the structural singularity and recover falsely deleted elements： the weak material element method and the tiny section element method. Three criteria （no structural singularity, no violated constraints and no change of structural weight） are introduced to judge iteration convergence. These criteria allow finding an appropriate threshold by adjusting a discount factor in the iteration procedure. To improve the efficiency, the original optimization model is transformed into a dual problem according to the dual theory and solved in its dual space. By using MSC/Nastran as the structural solver and MSC/Patran as the developing platform, a topological optimization software of frame structures is accomplished. Numerical examples show that the ICM method is very efficient for the topological optimization of frame structures.
