基于知识蒸馏和定位引导的Pointpillars点云检测网络

激光雷达数据由于其几何特性,被广泛应用于三维目标检测任务中。由于点云数据的稀疏性和不规则性,难以实现特征提取的质量和推理速度间的平衡。本文提出一种基于体柱特征编码的三维目标检测算法,以Pointpillars网络为基础,设计Teacher-Student模型框架对回归框尺度进行蒸馏,增加蒸馏损失,优化训练网络模型,提升特征提取的质量。为进一步提高模型检测效果,设计定位引导分类项,增加分类预测和回归预测之间的相关性,提高物体识别准确率。本网络所做改进没有引入额外的网络嵌入。算法在KITTI数据集上的实验结果表明,相比于基准网络,在三维模式下的平均精度值从60.65%提升到了64.69%,鸟瞰图模式下的平均精度值从67.74%提升到70.24%。模型推理速度为45 FPS,在提升检测精度的同时满足了实时性要求。

A Primal-Dual Infeasible-Interior-Point Algorithm for Multiple Objective Linear Programming Problems

A primal-dual infeasible interior point algorithm for multiple objective linear programming (MOLP) problems was presented. In contrast to the current MOLP algorithm. moving through the interior of polytope but not confining the iterates within the feasible region in our proposed algorithm result in a solution approach that is quite different and less sensitive to problem size, so providing the potential to dramatically improve the practical computation effectiveness.

Detection of Multiscale Center Point Objects Based on Parallel Network

Anchor-based detectors are widely used in object detection.To improve the accuracy of object detection,multiple anchor boxes are intensively placed on the input image,yet.Most of which are invalid.Although the anchor-free method can reduce the number of useless anchor boxes,the invalid ones still occupy a high proportion.On this basis,this paper proposes a multiscale center point object detection method based on parallel network to further reduce the number of useless anchor boxes.This study adopts the parallel network architecture of hourglass-104 and darknet-53 of which the first one outputs heatmaps to generate the center point for object feature location on the output attribute feature map of darknet-53.Combining feature pyramid and CIoU loss function,this algorithm is trained and tested on MSCOCO dataset,increasing the detection rate of target location and the accuracy rate of small object detection.Though resembling the state-of-the-art two-stage detectors in overall object detection accuracy,this algorithm is superior in speed.

基于PointCloudTransformer和优化集成学习的三维点云分类

针对三维点云的不规则性和无序性所导致的难于提取特征并进行分类的问题,提出了一种融合深度学习和集成学习的三维点云分类方法。首先,训练深度学习点云分类网络PointCloudTransformer,并使用主干网络提取点云特征,进而训练基分类器,获得基分类器集合;然后,针对集成学习算法设计基分类器选择模型,模型的优化目标为基分类器组合的差异性和平均总体精度。为了降低集成规模,本文基于增强后的白鲸优化算法提出了二元多目标白鲸优化算法,并使用该算法优化基分类器选择模型,获得集成剪枝方案集合;最后,采用多数投票法集成每个基分类器组合在测试集点云特征上的分类结果,获得最优基分类器组合,从而构建基于多目标优化剪枝的集成学习点云分类模型。在点云分类数据集上的实验结果表明,本文方法使用了更小的集成规模,获得了更高的集成精度,能够对多类别三维点云进行准确分类。

Development of vehicle-recognition method on water surfaces using LiDAR data:SPD^(2)(spherically stratified point projection with diameter and distance)

Swarm robot systems are an important application of autonomous unmanned surface vehicles on water surfaces.For monitoring natural environments and conducting security activities within a certain range using a surface vehicle,the swarm robot system is more efficient than the operation of a single object as the former can reduce cost and save time.It is necessary to detect adjacent surface obstacles robustly to operate a cluster of unmanned surface vehicles.For this purpose,a LiDAR(light detection and ranging)sensor is used as it can simultaneously obtain 3D information for all directions,relatively robustly and accurately,irrespective of the surrounding environmental conditions.Although the GPS(global-positioning-system)error range exists,obtaining measurements of the surface-vessel position can still ensure stability during platoon maneuvering.In this study,a three-layer convolutional neural network is applied to classify types of surface vehicles.The aim of this approach is to redefine the sparse 3D point cloud data as 2D image data with a connotative meaning and subsequently utilize this transformed data for object classification purposes.Hence,we have proposed a descriptor that converts the 3D point cloud data into 2D image data.To use this descriptor effectively,it is necessary to perform a clustering operation that separates the point clouds for each object.We developed voxel-based clustering for the point cloud clustering.Furthermore,using the descriptor,3D point cloud data can be converted into a 2D feature image,and the converted 2D image is provided as an input value to the network.We intend to verify the validity of the proposed 3D point cloud feature descriptor by using experimental data in the simulator.Furthermore,we explore the feasibility of real-time object classification within this framework.

Low-order Wavefront Error Compensation for Multi-field of Lithography Projection Objective Based on Interior Point Method

Low-order wavefront error account for a large proportion of wave aberrations.A compensation method for low order aberration of projection lithography objective based on Interior Point Method is presented.Compensation model between wavefront error and degree of movable lens freedom is established.Converting over-determined system to underdetermined system,the compensation is solved by Interior Point Method(IPM).The presented method is compared with direct solve the over-determined system.Then,other algorithm GA,EA and PS is compared with IPM.Simulation and experimental results show that the presented compensation method can obtained compensation with less residuals compared with direct solve the over-determined system.Also,the presented compensation method can reduce computation time and obtain results with less residuals compare with AGA,EA and PS.Moreover,after compensation,RMS of wavefront error of the experimental lithography projection objective decrease from 56.05 nm to 17.88 nm.

基于GCR-PointPillars的点云三维目标检测

针对PointPillars算法中存在识别与定位不准确的问题,提出一种GCR-PointPillars三维目标检测模型,该模型首先在Pillar特征网络中引入全局注意力机制,学习点云特征之间的相关性,增强伪图特征的全局信息交互能力;其次,基于ConvNeXt V2重新构建特征提取网络,提取更加丰富的语义信息,从而有效提升网络的学习能力;最后引入RDIoU来联合引导分类和回归任务,有效缓解分类和回归不一致的问题。文中模型在KITTI数据集中与基准网络相比,汽车类别在简单、中等、困难三种难度级别下分别提高了2.69%、4.29%、4.84%,并且推理速度达到25.8 f/s。实验结果表明,文中模型在保持实时性速度的同时,检测效果也有明显提升。

Compactness, Contractibility and Fixed Point Properties of the Pareto Sets in Multi-Objective Programming

This paper presents the Pareto solutions in continuous multi-objective mathematical programming. We discuss the role of some assumptions on the objective functions and feasible domain, the relationship between them, and compactness, contractibility and fixed point properties of the Pareto sets. The authors have tried to remove the concavity assumptions on the objective functions which are usually used in multi-objective maximization problems. The results are based on constructing a retraction from the feasible domain onto the Pareto-optimal set.

Optimal Test Points Selection Based on Multi-Objective Genetic Algorithm

A new approach to select anoptimal set of test points is proposed. The described method uses fault-wise table and multi-objective genetic algorithm to find the optimal set of test points. First, the fault-wise table is constructed whose entries are measurements associated with faults and test points. The selection of optimal test points is transformed to the selection of the columns that isolate the rows of the table. Then, four objectives are described according to practical test requirements. The multi-objective genetic algorithm is explained. Finally, the presented approach is illustrated by a practical example. The results indicate that the proposed method can efficiently and accurately find the optimal set of test points and is practical for large scale systems.

PVS-CNN:子流稀疏卷积优化的Point-Voxel CNN

针对经典的三维卷积网络在模型较大的场景上分割和检测的效率低和内存占用大的问题,提出PVS-CNN网络框架,通过更新哈希表和特征稀疏矩阵的方式实现了效率高且占用低的三维卷积,引用子流稀疏卷积改进PV-Conv。将PVS-CNN在ShapeNet和S3DIS数据集上进行评估,实验结果表明,所提出的PVS-CNN比PVCNN快3.6倍,GPU内存占用仅为PVCNN的0.55倍。在目标检测上,与F-PVCNN相比,PVS-CNN在时间效率和检测精度上全面优于F-PVCNN。

基于ObjectNess BING的海面多舰船目标检测

将一幅图像按照一个目标的大小进行缩放,然后计算其梯度特征,再对梯度特征进行标准化,二值化能够极大地提高目标候选区域的选择和检测计算效率,减少耗时。由于对海上舰船目标的检测是具有丰富角点的人造目标,对ObjectNess二值化标准梯度特征(binarized normed gradients,BING)方法中的目标候选区域提取算法进行改进,使其能够更加快速地进行候选区域的选择并保持较高的检测率。分析了海上多舰船目标的图像特征,提出了利用角点确定目标的候选基点,再利用ObjectNess BING检测模型训练获得的多目标尺寸进行候选区域的选择,对互联网上下载的多幅多舰船图像进行处理的结果表明,算法能够有效减少候选目标区域的数量并保持较高的检测概率。

基于改进人工势场法的避障路径规划研究

传统的人工势场法(APF)在路径规划领域因其简单性和高效性而被广泛采用,然而,这种方法往往会遇到局部最小值的问题,并且在动态环境中的适应性有限。为了解决这些问题,文中提出一种基于模拟退火算法(SA)改进的人工势场法。该改进方法结合人工势场法的实时避障能力和模拟退火法的全局优化特性,在所提出的改进方法中,通过在局部极小值附近添加随机目标点,使用模拟退火算法进行优化,从而有助于跳出局部最小值,并逐渐逼近全局最优或近似最优解。通过一系列的仿真实验表明,与传统人工势场法相比,基于模拟退火法的改进方法能够显著减少陷入局部最小值的情况,并在多种动态场景中表现出更强的鲁棒性和更优的路径规划效果。此外,该方法还展现了良好的实时性和适应性,能够满足车辆在复杂动态环境中进行避障和路径规划的需求。

基于目标检测和多视图几何的动态SLAM算法

目前大多数的SLAM系统主要针对静态场景,然而,在实际环境中不可避免地存在许多动态对象,这将大大降低算法的鲁棒性和相机的定位精度。针对动态对象造成的轨迹偏差问题,文中提出一种结合目标检测网络和多视图几何结构的动态SLAM算法。首先,基于YOLOv5算法框架,将骨干网络CSPDarkNet-53替换为轻量型L-FPN(Lite-FPN)结构,并使用VOC2007数据集进行预训练。与YOLOv5s原始模型相比,新网络的计算量减少了45.73%,检测速率提高了31.90%;然后,将检测物体划分为高动态对象、中动态对象以及低动态对象,利用多视图几何方法计算阈值,并根据阈值对中高动态对象进行二次检测,以决定是否剔除预测框中的特征点;最后,在TUM数据集上的实验结果显示,该方法在定位精度上平均提升了82.08%,证明了其在准确性方面的显著改进。

基于特征融合的改进型PointPillar点云目标检测

针对PointPillar在自动驾驶道路场景下对点云稀疏小目标检测效果差的问题,通过引入一种多尺度特征融合策略和注意力机制,提出一种点云目标检测网络Pillar-FFNet。针对网络中的特征提取问题,设计了一种基于残差结构的主干网络;针对馈入检测头的特征图没有充分利用高层特征的语义信息和低层特征的空间信息的问题,设计了一种简单有效的多尺度特征融合策略;针对主干网络提取的特征图中信息冗余的问题,提出了一种卷积注意力机制。为验证所提算法的性能,在KITTI和DAIR-V2X-I数据集上进行实验。实验结果表明,所提出的算法在KITTI数据集上与PointPillar相比,汽车、行人和骑行者的平均精度最大提高分别为0.84%,2.13%和4.02%;在DAIR-V2X-I数据集上与PointPillar相比,汽车、行人和骑行者的平均精度最大提高分别为0.33%,2.09%和4.71%,由此证明了所提方法对点云稀疏小目标检测的有效性。

改进PointNetLK的点云智能配准与位姿图优化方法

针对空间在轨服务任务中的非合作目标相对位姿测量问题,提出一种目标可测部位点云的智能配准方法。首先,通过Straight Through滤波算法对半物理仿真平台采集得到的点云进行目标提取,以消除背景数据等杂乱信息;其次,改进PointNetLK神经网络点云配准算法,将提取后的点云数据作为输入,从而获得初步配准结果,解决非合作目标先验信息缺失导致的无法配准问题;最后,建立基于位姿图的优化模型,以降低配准误差,提高配准精度。实验结果表明,与传统迭代最近点(ICP)算法相比,配准综合误差从6.3598降低到1.7291,精度提高约72.81%;单次耗时从33.16 s降低到4.2 s,效率提升约87.33%,与当前SM-ICP等其他算法相比,也具有一定的优势。

考虑能量转换装置可变效率的综合能源系统多目标优化调度

为了全面考虑综合能源系统的运行性能,提出一种计及能量转换装置可变效率的综合能源系统多目标优化运行方法。首先,对具有可变效率的能量转换设备运行特性曲线进行分段线性化处理。其次,建立以经济成本、碳排放量和综合能源利用效率为优化对象的多目标优化调度模型,并通过Charnes-Cooper变换解决综合能源利用效率目标的非凸非线性问题。最后,在求解该多目标模型时,运用规格化法平面约束法生成均匀分布的帕累托前沿,再采用改进熵权双基点法决策出折衷最优解。对修改的IEEE 33节点电网和8节点热网耦合的综合能源系统进行仿真分析。结果表明:优化后的模型和算法可以得到兼顾综合能源系统经济性、低碳性及高效性的优化运行方案,为决策者提供科学的决策参考。

基于多数据融合的机器人智能无序抓取系统开发

在机器人无序抓取场景中,现有的目标定位及位姿估计算法存在如鲁棒性不足、难以处理弱纹理和对称物体等问题,结合多数据融合方法开展机器人智能无序抓取系统开发。首先,针对复杂场景中的弱纹理目标,完成无序抓取系统的硬件设计。利用物理渲染合成样本增大数据集,并训练工件检测模型。通过Mask R-CNN网络进行目标识别,并设计了一种基于深度信息筛选的算法,用于简化堆叠场景;其次,提出基于点对特征匹配的位姿估计算法,利用FPFH算法估计点云特征,并通过RANSAC循环求解初始位姿,再使用ICP算法进行优化;最后,搭建基于视觉引导的机器人抓取系统,并在实验中验证了算法的快速、准确和稳定性。

基于注意力机制的PointPillars+三维目标检测

针对自动驾驶场景下的周边车辆及行人等目标的识别和定位问题,提出了一种基于注意力机制的PointPillars+三维目标检测算法.算法将完整空间按等尺寸柱均匀划分,从所有柱中提取内部点云的特征进而生成伪图;引入2种注意力机制,实现对伪图中特征信息的放大和抑制;使用卷积神经网络对注意力模块输出结果进一步处理,并使用SSD(single shot multibox detector)算法进行三维目标检测.结果表明:基于并行注意力机制的PointPillars+检测算法取得了良好的检测效果,相对于PointPillars算法,鸟瞰图下,平均均值精度(mAP)的中等难度均值(mAPm)从66.19增加到69.95,汽车的mAP从86.10增加到87.73;三维模式下,mAPm从59.20增加到62.55,汽车的mAP从74.99增加到76.25.

HESSIAN MATRIX BASED SADDLE POINT DETECTION FOR GRANULES SEGMENTATION IN 2D IMAGE

Segmenting the touching objects in an image has been remaining as a hot subject due to the problematic complexities, and a vast number of algorithms designed to tackle this issue have come into being since a decade ago. In this paper, a new granule segmentation algorithm is developed using saddle point as the cutting point. The image is binarized and then sequentially eroded to form a gray-scale topographic counterpart, followed by using Hessian matrix computation to search for the saddle point. The segmentation is performed by cutting through the saddle point and along the maximal gradient path on the topographic surface. The results of the algorithm test on the given real images indicate certain superiorities in both the segmenting robustness and execution time to the referenced methods.

基于knee points的改进多目标人工蜂群算法

传统的人工蜂群算法(Artificial Bee Colony algorithm,ABC)及其在多目标上的扩展(Multi Objective Artificial Bee Colony algorithm,MOABC)存在着在高维、多峰函数情况下收敛速度变慢、后期容易陷入局部最优以及寻优精度丢失等问题。基于knee points提高收敛性和分布性的特点,设计了一种快速识别knee point的算法并将其应用到多目标人工蜂群算法中,提出了一种基于knee points的改进多目标人工蜂群算法(Kn MOABC)。算法在迭代过程中考虑pareto支配关系的同时,优先选择knee point作为下一代个体,极大地增强了算法的收敛速度,同时,在knee point识别算法中加入自适应的策略以保持良好的分布性。实验结果表明,Kn MOABC的性能优于三个最新的多目标人工蜂群对比算法。