激光雷达点云3D物体检测,对于小物体如行人、自行车的检测精度较低,容易漏检误检,提出一种多尺度Transformer激光雷达点云3D物体检测方法 MSPT-RCNN(multi-scale point transformer-RCNN),提高点云3D物体检测精度。该方法包含两个阶段,...激光雷达点云3D物体检测,对于小物体如行人、自行车的检测精度较低,容易漏检误检,提出一种多尺度Transformer激光雷达点云3D物体检测方法 MSPT-RCNN(multi-scale point transformer-RCNN),提高点云3D物体检测精度。该方法包含两个阶段,即第一阶段(RPN)和第二阶段(RCNN)。RPN阶段通过多尺度Transformer网络提取点云特征,该网络包含多尺度邻域嵌入模块和跳跃连接偏移注意力模块,获取多尺度邻域几何信息和不同层次全局语义信息,生成高质量初始3D包围盒;在RCNN阶段,引入包围盒内的点云多尺度邻域几何信息,优化了包围盒位置、尺寸、朝向和置信度等信息。实验结果表明,该方法(MSPT-RCNN)具有较高检测精度,特别是对于远处和较小物体,提升更高。MSPT-RCNN通过有效学习点云数据中的多尺度几何信息,提取不同层次有效的语义信息,能够有效提升3D物体检测精度。展开更多
3D物体检测是计算机视觉的一个重要研究方向,在自动驾驶等领域有着广泛的应用.现有的前沿工作采用端到端的深度学习方法,虽然达到了很好的检测效果但存在着算法复杂度高、计算量大、实时性不够等问题.经过分析发现3D物体检测中的“部分...3D物体检测是计算机视觉的一个重要研究方向,在自动驾驶等领域有着广泛的应用.现有的前沿工作采用端到端的深度学习方法,虽然达到了很好的检测效果但存在着算法复杂度高、计算量大、实时性不够等问题.经过分析发现3D物体检测中的“部分任务”并不适合使用深度学习的方法进行解决,为此提出了一种基于异构方法的3D物体检测方法,该方法在检测过程中同时使用深度学习和传统算法,将检测过程划分为多任务阶段:1)利用深度学习方法从被检测图片中获取被检测物体的mask、物体类别等信息;2)基于mask,利用快速聚类方法从雷达点云空间中筛选出目标物体的表面雷达点;3)利用物体mask、类别、雷达点云等信息计算物体朝向、边框等信息,最终实现3D物体检测.对该方法进行了系统实现,称之为HA3D(a heterogeneous approach for 3D object detection).经实验表明:在针对汽车的3D检测数据集KITTI上,该方法与代表性的基于深度学习的3D物体检测方法相比,在检测精度下降接受范围内(2.0%),速度提升了52.2%,精确率与计算时间的比值提升了49%.从综合表现上来看,方法具有明显的优势.展开更多
This paper presents a complete system for scanning the geometry and texture of a large 3D object, then the automatic registration is performed to obtain a whole realistic 3D model. This system is composed of one line ...This paper presents a complete system for scanning the geometry and texture of a large 3D object, then the automatic registration is performed to obtain a whole realistic 3D model. This system is composed of one line strip laser and one color CCD camera. The scanned object is pictured twice by a color CCD camera. First, the texture of the scanned object is taken by a color CCD camera. Then the 3D information of the scanned object is obtained from laser plane equations. This paper presents a practical way to implement the three dimensional measuring method and the automatic registration of a large 3D object and a pretty good result is obtained after experiment verification.展开更多
In this paper, we propose a new algorithm for temporally consistent depth map estimation to generate three-dimensional video. The proposed algorithm adaptively computes the matching cost using a temporal weighting fun...In this paper, we propose a new algorithm for temporally consistent depth map estimation to generate three-dimensional video. The proposed algorithm adaptively computes the matching cost using a temporal weighting function, which is obtained by block-based moving object detection and motion estimation with variable block sizes. Experimental results show that the proposed algorithm improves the temporal consistency of the depth video and reduces by about 38% both the flickering artefact in the synthesized view and the number of coding bits for depth video coding.展开更多
Complementary metal oxide semiconductor(CMOS) image sensors(CIS) are being widely used in digital video cameras, web cameras, digital single lens reflex camera(DSLR), smart phones and so on, owing to their high level ...Complementary metal oxide semiconductor(CMOS) image sensors(CIS) are being widely used in digital video cameras, web cameras, digital single lens reflex camera(DSLR), smart phones and so on, owing to their high level of integration, random accessibility, and low-power operation. It needs to be installed with the cover glass in practical applications to protect the sensor from damage, mechanical issues,and environmental conditions, which, however, limits the accuracy and usability of the sensor due to the reflection in the optical path from air-to-cover glass-to-air. In this work, the flexible 3D nanocone anti-reflection(AR) film with controlled aspect ratio was firstly employed to reduce the light reflection at air/cover glass/air interfaces by directly attaching onto the front and rear sides of the CIS cover glass.As both the front and rear sides of cover glass were coated by the AR film, the output image quality was found to be improved with external quantum efficiency increased by 7%, compared with that without AR film. The mean digital data value, root-mean-square contrast, and dynamic range are increased by45.14%, 38.61% and 57, respectively, for the output image with AR films. These results provide a novel and facile pathway to improve the CIS performance and also could be extended to rational design of other image sensors and optoelectronic devices.展开更多
文摘激光雷达点云3D物体检测,对于小物体如行人、自行车的检测精度较低,容易漏检误检,提出一种多尺度Transformer激光雷达点云3D物体检测方法 MSPT-RCNN(multi-scale point transformer-RCNN),提高点云3D物体检测精度。该方法包含两个阶段,即第一阶段(RPN)和第二阶段(RCNN)。RPN阶段通过多尺度Transformer网络提取点云特征,该网络包含多尺度邻域嵌入模块和跳跃连接偏移注意力模块,获取多尺度邻域几何信息和不同层次全局语义信息,生成高质量初始3D包围盒;在RCNN阶段,引入包围盒内的点云多尺度邻域几何信息,优化了包围盒位置、尺寸、朝向和置信度等信息。实验结果表明,该方法(MSPT-RCNN)具有较高检测精度,特别是对于远处和较小物体,提升更高。MSPT-RCNN通过有效学习点云数据中的多尺度几何信息,提取不同层次有效的语义信息,能够有效提升3D物体检测精度。
文摘3D物体检测是计算机视觉的一个重要研究方向,在自动驾驶等领域有着广泛的应用.现有的前沿工作采用端到端的深度学习方法,虽然达到了很好的检测效果但存在着算法复杂度高、计算量大、实时性不够等问题.经过分析发现3D物体检测中的“部分任务”并不适合使用深度学习的方法进行解决,为此提出了一种基于异构方法的3D物体检测方法,该方法在检测过程中同时使用深度学习和传统算法,将检测过程划分为多任务阶段:1)利用深度学习方法从被检测图片中获取被检测物体的mask、物体类别等信息;2)基于mask,利用快速聚类方法从雷达点云空间中筛选出目标物体的表面雷达点;3)利用物体mask、类别、雷达点云等信息计算物体朝向、边框等信息,最终实现3D物体检测.对该方法进行了系统实现,称之为HA3D(a heterogeneous approach for 3D object detection).经实验表明:在针对汽车的3D检测数据集KITTI上,该方法与代表性的基于深度学习的3D物体检测方法相比,在检测精度下降接受范围内(2.0%),速度提升了52.2%,精确率与计算时间的比值提升了49%.从综合表现上来看,方法具有明显的优势.
文摘This paper presents a complete system for scanning the geometry and texture of a large 3D object, then the automatic registration is performed to obtain a whole realistic 3D model. This system is composed of one line strip laser and one color CCD camera. The scanned object is pictured twice by a color CCD camera. First, the texture of the scanned object is taken by a color CCD camera. Then the 3D information of the scanned object is obtained from laser plane equations. This paper presents a practical way to implement the three dimensional measuring method and the automatic registration of a large 3D object and a pretty good result is obtained after experiment verification.
基金supported by the National Research Foundation of Korea Grant funded by the Korea Ministry of Science and Technology under Grant No. 2012-0009228
文摘In this paper, we propose a new algorithm for temporally consistent depth map estimation to generate three-dimensional video. The proposed algorithm adaptively computes the matching cost using a temporal weighting function, which is obtained by block-based moving object detection and motion estimation with variable block sizes. Experimental results show that the proposed algorithm improves the temporal consistency of the depth video and reduces by about 38% both the flickering artefact in the synthesized view and the number of coding bits for depth video coding.
基金financially supported by the National Natural Science Foundation of China(61474128,21503261,61504155and 61404145)Youth Innovation Fund for Interdisciplinary Research of SARI(Y526453233,141004)+2 种基金Science & Technology Commission of Shanghai Municipality(14JC1492900,14511102302,15DZ1100502)Youth Innovation Promotion Association,CAS(2013302)Development Fund for Information communication and integrated circuit technology public service platform(No.2016-14)supported by Zhangjiang Adminstrative Committee
文摘Complementary metal oxide semiconductor(CMOS) image sensors(CIS) are being widely used in digital video cameras, web cameras, digital single lens reflex camera(DSLR), smart phones and so on, owing to their high level of integration, random accessibility, and low-power operation. It needs to be installed with the cover glass in practical applications to protect the sensor from damage, mechanical issues,and environmental conditions, which, however, limits the accuracy and usability of the sensor due to the reflection in the optical path from air-to-cover glass-to-air. In this work, the flexible 3D nanocone anti-reflection(AR) film with controlled aspect ratio was firstly employed to reduce the light reflection at air/cover glass/air interfaces by directly attaching onto the front and rear sides of the CIS cover glass.As both the front and rear sides of cover glass were coated by the AR film, the output image quality was found to be improved with external quantum efficiency increased by 7%, compared with that without AR film. The mean digital data value, root-mean-square contrast, and dynamic range are increased by45.14%, 38.61% and 57, respectively, for the output image with AR films. These results provide a novel and facile pathway to improve the CIS performance and also could be extended to rational design of other image sensors and optoelectronic devices.
