人工智能培训

搜索

人工智能论文:在LIDAR上使用2D-3D SIAMESE网络的高效跟踪建议(Efficient Tracking Proposals using 2D-3D

[复制链接]
comceo 发表于 2019-3-26 11:43:42 | 显示全部楼层 |阅读模式
comceo 2019-3-26 11:43:42 339 0 显示全部楼层
人工智能论文:在LIDAR上使用2D-3D SIAMESE网络的高效跟踪建议(Efficient Tracking Proposals using 2D-3D Siamese Networks on LIDAR)由于数据的稀疏性和密集的搜索空间,在LIDAR点云中跟踪车辆是一项具有挑战性的任务。 pointcloud中缺乏结构阻碍了通常在2D对象跟踪中使用的卷积和相关滤波器的使用。此外,构造点云是麻烦的,并且意味着丢失细粒度信息。结果,在3D空间中生成提议是昂贵且低效的。在本文中,我们利用LIDAR点云的密集和结构化鸟瞰图(BEV)表示来有效地搜索感兴趣的对象。我们使用有效的Region ProposalNetwork并在3D中生成少量对象提案。接下来,通过利用3D Siamese网络的相似性,我们选择了三维候选对象。我们将后者3D Siamese网络规范化以完成形状,以增强其辨别能力。我们的方法试图解决BEV空间中的高效搜索空间和使用3D LIDAR点云的有意义的选择。我们证明BEV中的RegionProposal优于贝叶斯方法,如Kalman和ParticleFilters,提供了大幅度的提议,并且这些候选者适用于3D Siamese网络。通过端对端训练我们的方法,当仅使用16个候选者时,我们在Successand Precision中将车辆跟踪的先前基线表现为12%/ 18%。
Tracking vehicles in LIDAR point clouds is a challenging task due to thesparsity of the data and the dense search space.The lack of structure in pointclouds impedes the use of convolution and correlation filters usually employedin 2D object tracking.In addition, structuring point clouds is cumbersome andimplies losing fine-grained information.As a result, generating proposals in3D space is expensive and inefficient.In this paper, we leverage the dense andstructured Bird Eye View (BEV) representation of LIDAR point clouds toefficiently search for objects of interest.We use an efficient Region ProposalNetwork and generate a small number of object proposals in 3D.Successively, werefine our selection of 3D object candidates by exploiting the similaritycapability of a 3D Siamese network.We regularize the latter 3D Siamese networkfor shape completion to enhance its discrimination capability.Our methodattempts to solve both for an efficient search space in the BEV space and ameaningful selection using 3D LIDAR point cloud.We show that the RegionProposal in the BEV outperforms Bayesian methods such as Kalman and ParticleFilters in providing proposal by a significant margin and that such candidatesare suitable for the 3D Siamese network.By training our method end-to-end, weoutperform the previous baseline in vehicle tracking by 12% / 18% in Successand Precision when using only 16 candidates.人工智能论文:在LIDAR上使用2D-3D SIAMESE网络的高效跟踪建议(Efficient Tracking Proposals using 2D-3D Siamese Networks on LIDAR) tSNw0SZ60Sv1nK11.jpg
URL地址:https://arxiv.org/abs/1903.10168     ----pdf下载地址:https://arxiv.org/pdf/1903.10168    ----人工智能论文:在LIDAR上使用2D-3D SIAMESE网络的高效跟踪建议(Efficient Tracking Proposals using 2D-3D Siamese Networks on LIDAR)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

comceo当前离线
新手上路

查看:339 | 回复:0

快速回复 返回顶部 返回列表