目前,ToF(Time of Flight)三维成像技术在人脸检测、3D目标识别、三维重建等视觉任务领域具有广阔的应用前景。然而,用ToF相机所获得的深度信息往往存在与像素、温度、深度畸变、多径干扰以及背景光相关的噪声干扰。现有的ToF优化算法...目前,ToF(Time of Flight)三维成像技术在人脸检测、3D目标识别、三维重建等视觉任务领域具有广阔的应用前景。然而,用ToF相机所获得的深度信息往往存在与像素、温度、深度畸变、多径干扰以及背景光相关的噪声干扰。现有的ToF优化算法耗时较大且很难保留目标的细节信息,这些问题严重影响了ToF相机的实际应用。针对以上问题,本文提出一种实时的基于振幅图的ToF深度图优化方法。首先通过ToF接收端采集的原始数据生成带有噪声的振幅图像。针对振幅图中的噪声,选用快速高效的双边网格滤波对振幅图进行去噪。然后,利用优化后的振幅图生成掩码以分割出深度图中前景和背景区域。同时,对深度图中的噪声以及误差像素用滤波的方式优化,最后将优化后的深度图和掩码融合生成最终的深度图。实验结果表明,本文所提算法可以实时有效地滤除深度图噪声,去除背景噪声的干扰,同时能很好地保留深度图中目标对象的细节信息。有助于ToF相机拥有更广泛的应用场景。展开更多
In order to study the possibility of improving the timing performance of the time of flight （TOF） systems, which are made of plastic scintillator counters, and read out by photomultiplier tubes （PMT） with mesh dyn...In order to study the possibility of improving the timing performance of the time of flight （TOF） systems, which are made of plastic scintillator counters, and read out by photomultiplier tubes （PMT） with mesh dynodes and conventional electronics, we have conducted a study using faster PMTs and ultra fast waveform digitizers to read out the plastic scintillators. Different waveform analysis methods are used to calculate the time resolution of such a system. Results are compared with the conventional discriminating method based on a threshold and pulse height. Our tests and analysis show that significant timing performance improvements can be achieved by using this new system.展开更多
To increase spatial resolution and signal-to-noise ratio in PET imaging,we present in this paper the design and performance evaluation of a PET detector module combining both depth-of-interaction(DOI) and time-offligh...To increase spatial resolution and signal-to-noise ratio in PET imaging,we present in this paper the design and performance evaluation of a PET detector module combining both depth-of-interaction(DOI) and time-offlight(TOF) capabilities.The detector module consists of a staggered dual-layer LYSO block with2 mm × 2 mm × 7 mm crystals.MR-compatible SiPM sensors(MicroFJ-30035-TSV,SensL) are assembled into an 8× 8 array.SiPM signals from both fast and slow outputs are read out by a 128-channel ASIC chip.To test its performance,a flood histogram is acquired with a ^(22)Na point source on top of the detector,and the energy resolution and the coincidence resolving time(CRT) value for each individual crystal are measured.The flood histogram shows excellent crystal separation in both layers.The average energy resolution at 511 keV is 14.0 and 12.7%at the bottom and top layers,respectively.The average CRT of a single crystal is 635 and 565 ps at the bottom and top layers,respectively.In conclusion,the compact DOI-TOF PET detector module is of excellent crystal identification capability,good energy resolution and reasonable time resolution and has promising application prospective in clinical TOF PET,PET/MRI,and brain PET systems.展开更多
针对工业生产线的堆叠矩形物品的识别与分拣问题,设计了一套由TOF(time of flight)相机、并联机器人与夹具等组合而成的产品分拣系统。采用一种基于深度图与RGB图结合的三维矩形检测算法,实现堆叠物品的识别与空间位姿的计算。运用机器...针对工业生产线的堆叠矩形物品的识别与分拣问题,设计了一套由TOF(time of flight)相机、并联机器人与夹具等组合而成的产品分拣系统。采用一种基于深度图与RGB图结合的三维矩形检测算法,实现堆叠物品的识别与空间位姿的计算。运用机器人定位,通过简易的手眼标定法,对算法进行验证。最后,结合视觉系统与机器人控制系统进行抓取测试。通过不同数量下反复抓取的实验,检测该系统的分拣能力。展开更多
文摘目前,ToF(Time of Flight)三维成像技术在人脸检测、3D目标识别、三维重建等视觉任务领域具有广阔的应用前景。然而,用ToF相机所获得的深度信息往往存在与像素、温度、深度畸变、多径干扰以及背景光相关的噪声干扰。现有的ToF优化算法耗时较大且很难保留目标的细节信息,这些问题严重影响了ToF相机的实际应用。针对以上问题,本文提出一种实时的基于振幅图的ToF深度图优化方法。首先通过ToF接收端采集的原始数据生成带有噪声的振幅图像。针对振幅图中的噪声,选用快速高效的双边网格滤波对振幅图进行去噪。然后,利用优化后的振幅图生成掩码以分割出深度图中前景和背景区域。同时,对深度图中的噪声以及误差像素用滤波的方式优化,最后将优化后的深度图和掩码融合生成最终的深度图。实验结果表明,本文所提算法可以实时有效地滤除深度图噪声,去除背景噪声的干扰,同时能很好地保留深度图中目标对象的细节信息。有助于ToF相机拥有更广泛的应用场景。
基金Supported by National Natural Science Foundation of China(10979003)Main Direction Program of Knowledge Innovation Project of Chinese Academy of Sciences
文摘In order to study the possibility of improving the timing performance of the time of flight （TOF） systems, which are made of plastic scintillator counters, and read out by photomultiplier tubes （PMT） with mesh dynodes and conventional electronics, we have conducted a study using faster PMTs and ultra fast waveform digitizers to read out the plastic scintillators. Different waveform analysis methods are used to calculate the time resolution of such a system. Results are compared with the conventional discriminating method based on a threshold and pulse height. Our tests and analysis show that significant timing performance improvements can be achieved by using this new system.
基金supported in part by Fundamental Research Funds for the Central Universities(No.FRF-TP-15-114A1)National Natural Science Foundation of China(Nos.11375096,11505300)Tsinghua University Initiative Scientific Research Program(No.20131089289)
文摘To increase spatial resolution and signal-to-noise ratio in PET imaging,we present in this paper the design and performance evaluation of a PET detector module combining both depth-of-interaction(DOI) and time-offlight(TOF) capabilities.The detector module consists of a staggered dual-layer LYSO block with2 mm × 2 mm × 7 mm crystals.MR-compatible SiPM sensors(MicroFJ-30035-TSV,SensL) are assembled into an 8× 8 array.SiPM signals from both fast and slow outputs are read out by a 128-channel ASIC chip.To test its performance,a flood histogram is acquired with a ^(22)Na point source on top of the detector,and the energy resolution and the coincidence resolving time(CRT) value for each individual crystal are measured.The flood histogram shows excellent crystal separation in both layers.The average energy resolution at 511 keV is 14.0 and 12.7%at the bottom and top layers,respectively.The average CRT of a single crystal is 635 and 565 ps at the bottom and top layers,respectively.In conclusion,the compact DOI-TOF PET detector module is of excellent crystal identification capability,good energy resolution and reasonable time resolution and has promising application prospective in clinical TOF PET,PET/MRI,and brain PET systems.
文摘针对工业生产线的堆叠矩形物品的识别与分拣问题,设计了一套由TOF(time of flight)相机、并联机器人与夹具等组合而成的产品分拣系统。采用一种基于深度图与RGB图结合的三维矩形检测算法,实现堆叠物品的识别与空间位姿的计算。运用机器人定位,通过简易的手眼标定法,对算法进行验证。最后,结合视觉系统与机器人控制系统进行抓取测试。通过不同数量下反复抓取的实验,检测该系统的分拣能力。
