目的视觉感知技术是智能车系统中的一项关键技术,但是在复杂挑战下如何有效提高视觉性能已经成为智能驾驶领域的重要研究内容。本文将人工社会(artificial societies)、计算实验(computational experiments)和平行执行(parallel executi...目的视觉感知技术是智能车系统中的一项关键技术,但是在复杂挑战下如何有效提高视觉性能已经成为智能驾驶领域的重要研究内容。本文将人工社会(artificial societies)、计算实验(computational experiments)和平行执行(parallel execution)构成的ACP方法引入智能驾驶的视觉感知领域,提出了面向智能驾驶的平行视觉感知,解决了视觉模型合理训练和评估问题,有助于智能驾驶进一步走向实际应用。方法平行视觉感知通过人工子系统组合来模拟实际驾驶场景,构建人工驾驶场景使之成为智能车视觉感知的"计算实验室";借助计算实验两种操作模式完成视觉模型训练与评估;最后采用平行执行动态优化视觉模型,保障智能驾驶对复杂挑战的感知与理解长期有效。结果实验表明,目标检测的训练阶段虚实混合数据最高精度可达60.9%,比单纯用KPC(包括:KITTI(Karlsruhe Institute of Technology and Toyota Technological Institute),PASCAL VOC(pattern analysis,statistical modelling and computational learning visual object classes)和MS COCO(Microsoft common objects in context))数据和虚拟数据分别高出17.9%和5.3%;在评估阶段相较于基准数据,常规任务(-30°且垂直移动)平均精度下降11.3%,环境任务(雾天)平均精度下降21.0%,困难任务(所有挑战)平均精度下降33.7%。结论本文为智能驾驶设计和实施了在实际驾驶场景难以甚至无法进行的视觉计算实验,对复杂视觉挑战进行分析和评估,具备加强智能车在行驶过程中感知和理解周围场景的意义。展开更多
目的受遮挡与累积误差因素的影响,现有目标6维(6 dimensions,6D)姿态实时追踪方法在复杂场景中表现不佳。为此,提出了一种高鲁棒性的刚体目标6D姿态实时追踪网络。方法在网络的整体设计上,将当前帧彩色图像和深度图像(red green blue-de...目的受遮挡与累积误差因素的影响,现有目标6维(6 dimensions,6D)姿态实时追踪方法在复杂场景中表现不佳。为此,提出了一种高鲁棒性的刚体目标6D姿态实时追踪网络。方法在网络的整体设计上,将当前帧彩色图像和深度图像(red green blue-depth map,RGB-D)与前一帧姿态估计结果经升维残差采样滤波和特征编码处理获得姿态差异,与前一帧姿态估计结果共同计算目标当前的6D姿态;在残差采样滤波模块的设计中,采用自门控swish(searching for activation functions)激活函数保留目标细节特征,提高目标姿态追踪的准确性;在特征聚合模块的设计中,将提取的特征分解为水平与垂直两个方向分量,分别从时间和空间上捕获长程依赖并保留位置信息,生成一组具有位置与时间感知的互补特征图,加强目标特征提取能力,从而加速网络收敛。结果实验选用YCBVideo(Yale-CMU-Berkeley-video)和YCBInEoAT(Yale-CMU-Berkeley in end-of-arm-tooling)数据集。实验结果表明,本文方法追踪速度达到90.9 Hz,追踪精度模型点平均距离(average distance of model points,ADD)和最近点的平均距离(average closest point distance,ADD-S)分别达到93.24及95.84,均高于同类相关方法。本文方法的追踪精度指标ADD和ADD-S在追踪精度和追踪速度上均领先于目前其他的刚体姿态追踪方法,与se(3)-TrackNet网络相比,本文方法在6000组少量合成数据训练的条件下分别高出25.95和30.91,在8000组少量合成数据训练的条件下分别高出31.72和28.75,在10000组少量合成数据训练的条件下分别高出35.57和21.07,且在严重遮挡场景下能够实现对目标的高鲁棒6D姿态追踪。结论本文网络在合成数据驱动条件下,可以更好地完成实时准确追踪目标6D姿态,网络收敛速度快,实验结果验证了本文方法的有效性。展开更多
In order to solve the problems of small monitoring range,long time and high cost of existing sedimentation observation methods,based on two-view sentinel No.1 radar images of Guqiao mining area in Huainan City from No...In order to solve the problems of small monitoring range,long time and high cost of existing sedimentation observation methods,based on two-view sentinel No.1 radar images of Guqiao mining area in Huainan City from November 4,2017 to November 28,2017,surface change information was obtained in combination with D-InSAR,and the three-dimensional surface deformation was monitored by two-pass method and single line of sight D-InSAR method.The results show that during the research period of 24 d,the maximum deformation of the mining area reached 71 mm,and the southern subsidence was the most obvious,which was in line with the mining subsidence law.The maximum displacement from the north to the south was about 250 mm,while the maximum displacement from the east to the west was about 80 mm,and the maximum subsidence in the center was 110 mm.It is concluded that D-InSAR technique has a good effect on the inversion of the mining subsidence,and this method is suitable for three-dimensional surface monitoring in areas with similar geological conditions.The monitoring results have certain reference value.展开更多
Depth-image-based rendering(DIBR) is widely used in 3 DTV, free-viewpoint video, and interactive 3 D graphics applications. Typically, synthetic images generated by DIBR-based systems incorporate various distortions, ...Depth-image-based rendering(DIBR) is widely used in 3 DTV, free-viewpoint video, and interactive 3 D graphics applications. Typically, synthetic images generated by DIBR-based systems incorporate various distortions, particularly geometric distortions induced by object dis-occlusion. Ensuring the quality of synthetic images is critical to maintaining adequate system service. However, traditional 2 D image quality metrics are ineffective for evaluating synthetic images as they are not sensitive to geometric distortion. In this paper, we propose a novel no-reference image quality assessment method for synthetic images based on convolutional neural networks, introducing local image saliency as prediction weights. Due to the lack of existing training data, we construct a new DIBR synthetic image dataset as part of our contribution. Experiments were conducted on both the public benchmark IRCCyN/IVC DIBR image dataset and our own dataset. Results demonstrate that our proposed metric outperforms traditional 2 D image quality metrics and state-of-the-art DIBR-related metrics.展开更多
SAR images commonly suffer fromspeckle noise,posing a significant challenge in their analysis and interpretation.Existing convolutional neural network(CNN)based despeckling methods have shown great performance in remo...SAR images commonly suffer fromspeckle noise,posing a significant challenge in their analysis and interpretation.Existing convolutional neural network(CNN)based despeckling methods have shown great performance in removing speckle noise.However,these CNN-basedmethods have a fewlimitations.They do not decouple complex background information in amulti-resolutionmanner.Moreover,they have deep network structures thatmay result in many parameters,limiting their applicability tomobile devices.Furthermore,extracting key speckle information in the presence of complex background is also a major problem with SAR.The proposed study addresses these limitations by introducing a lightweight pyramid and attention-based despeckling(PAN-Despeck)network.The primary objective is to enhance image quality and enable improved information interpretation,particularly on mobile devices and scenarios involving complex backgrounds.The PAN-Despeck network leverages domainspecific knowledge and integrates Gaussian Laplacian image pyramid decomposition for multi-resolution image analysis.By utilizing this approach,complex background information can be effectively decoupled,leading to enhanced despeckling performance.Furthermore,the attention mechanism selectively focuses on key speckle features and facilitates complex background removal.The network incorporates recursive and residual blocks to ensure computational efficiency and accelerate training speed,making it lightweight while maintaining high performance.Through comprehensive evaluations,it is demonstrated that PAN-Despeck outperforms existing image restoration methods.With an impressive average peak signal-to-noise ratio(PSNR)of 28.355114 and a remarkable structural similarity index(SSIM)of 0.905467,it demonstrates exceptional performance in effectively reducing speckle noise in SAR images.The source code for the PAN-DeSpeck network is available on GitHub.展开更多
The traditional Range Doppler(RD)algorithm is unable to meet practical needs owing to the limit of resolution.The order of fractional Fourier Transform(FrFT)and the length of sampling signals affect SAR imaging perfor...The traditional Range Doppler(RD)algorithm is unable to meet practical needs owing to the limit of resolution.The order of fractional Fourier Transform(FrFT)and the length of sampling signals affect SAR imaging performance when FrFT is applied to RD algorithm.To overcome the above shortcomings,the purpose of this paper is to propose a high-resolution SAR image algorithm by using the optimal order of FrFT and the sample length constraints for the range direction.The expression of the optimal order of SAR range signals via FrFT is deduced in detail.The initial sample length and its constraints are proposed to obtain the best sample length of SAR range signals.Experimental results demonstrate that,when the range sampling-length changes in a certain interval,the best sampling-length will be obtained,which the best values of the range resolution,PSLR and ISLR,will be derived respectively.Compared with traditional RD algorithm,the main-lobe width of the peak-point target of the proposed algorithm is narrow in the range direction.While the peak amplitude of the first side-lobe is reduced significantly,those of other side-lobes also drop in various degrees.展开更多
There is difficulty for distinguishing of river and shadow in Synthetic Aperture Radar (SAR) images. A method of river segmentation in SAR images based on wavelet energy and gradient is proposed in this paper. It main...There is difficulty for distinguishing of river and shadow in Synthetic Aperture Radar (SAR) images. A method of river segmentation in SAR images based on wavelet energy and gradient is proposed in this paper. It mainly includes two algorithms: coarse segmentation and refined segmen- tation. Firstly, The river regions are coarsely segmented by the wavelet energy feature,and then refined segmented accurately by the gradient threshold which is got adaptively. The experimental results show the validity of the method, which provides a good foundation for targets detection above the river.展开更多
文摘目的视觉感知技术是智能车系统中的一项关键技术,但是在复杂挑战下如何有效提高视觉性能已经成为智能驾驶领域的重要研究内容。本文将人工社会(artificial societies)、计算实验(computational experiments)和平行执行(parallel execution)构成的ACP方法引入智能驾驶的视觉感知领域,提出了面向智能驾驶的平行视觉感知,解决了视觉模型合理训练和评估问题,有助于智能驾驶进一步走向实际应用。方法平行视觉感知通过人工子系统组合来模拟实际驾驶场景,构建人工驾驶场景使之成为智能车视觉感知的"计算实验室";借助计算实验两种操作模式完成视觉模型训练与评估;最后采用平行执行动态优化视觉模型,保障智能驾驶对复杂挑战的感知与理解长期有效。结果实验表明,目标检测的训练阶段虚实混合数据最高精度可达60.9%,比单纯用KPC(包括:KITTI(Karlsruhe Institute of Technology and Toyota Technological Institute),PASCAL VOC(pattern analysis,statistical modelling and computational learning visual object classes)和MS COCO(Microsoft common objects in context))数据和虚拟数据分别高出17.9%和5.3%;在评估阶段相较于基准数据,常规任务(-30°且垂直移动)平均精度下降11.3%,环境任务(雾天)平均精度下降21.0%,困难任务(所有挑战)平均精度下降33.7%。结论本文为智能驾驶设计和实施了在实际驾驶场景难以甚至无法进行的视觉计算实验,对复杂视觉挑战进行分析和评估,具备加强智能车在行驶过程中感知和理解周围场景的意义。
文摘目的受遮挡与累积误差因素的影响,现有目标6维(6 dimensions,6D)姿态实时追踪方法在复杂场景中表现不佳。为此,提出了一种高鲁棒性的刚体目标6D姿态实时追踪网络。方法在网络的整体设计上,将当前帧彩色图像和深度图像(red green blue-depth map,RGB-D)与前一帧姿态估计结果经升维残差采样滤波和特征编码处理获得姿态差异,与前一帧姿态估计结果共同计算目标当前的6D姿态;在残差采样滤波模块的设计中,采用自门控swish(searching for activation functions)激活函数保留目标细节特征,提高目标姿态追踪的准确性;在特征聚合模块的设计中,将提取的特征分解为水平与垂直两个方向分量,分别从时间和空间上捕获长程依赖并保留位置信息,生成一组具有位置与时间感知的互补特征图,加强目标特征提取能力,从而加速网络收敛。结果实验选用YCBVideo(Yale-CMU-Berkeley-video)和YCBInEoAT(Yale-CMU-Berkeley in end-of-arm-tooling)数据集。实验结果表明,本文方法追踪速度达到90.9 Hz,追踪精度模型点平均距离(average distance of model points,ADD)和最近点的平均距离(average closest point distance,ADD-S)分别达到93.24及95.84,均高于同类相关方法。本文方法的追踪精度指标ADD和ADD-S在追踪精度和追踪速度上均领先于目前其他的刚体姿态追踪方法,与se(3)-TrackNet网络相比,本文方法在6000组少量合成数据训练的条件下分别高出25.95和30.91,在8000组少量合成数据训练的条件下分别高出31.72和28.75,在10000组少量合成数据训练的条件下分别高出35.57和21.07,且在严重遮挡场景下能够实现对目标的高鲁棒6D姿态追踪。结论本文网络在合成数据驱动条件下,可以更好地完成实时准确追踪目标6D姿态,网络收敛速度快,实验结果验证了本文方法的有效性。
基金the Talent Introduction Project of Anhui University of Science and Technology(ZHYJ202104)Horizontal Cooperation Project(881079,880554,880982)Innovation and Entrepreneurship Project of National College Students(S202310879289,S202310879296,X202310879098,X20231087-9097).
文摘In order to solve the problems of small monitoring range,long time and high cost of existing sedimentation observation methods,based on two-view sentinel No.1 radar images of Guqiao mining area in Huainan City from November 4,2017 to November 28,2017,surface change information was obtained in combination with D-InSAR,and the three-dimensional surface deformation was monitored by two-pass method and single line of sight D-InSAR method.The results show that during the research period of 24 d,the maximum deformation of the mining area reached 71 mm,and the southern subsidence was the most obvious,which was in line with the mining subsidence law.The maximum displacement from the north to the south was about 250 mm,while the maximum displacement from the east to the west was about 80 mm,and the maximum subsidence in the center was 110 mm.It is concluded that D-InSAR technique has a good effect on the inversion of the mining subsidence,and this method is suitable for three-dimensional surface monitoring in areas with similar geological conditions.The monitoring results have certain reference value.
基金sponsored by the National Key R&D Program of China (No. 2017YFB1002702)the National Natural Science Foundation of China (Nos. 61572058, 61472363)
文摘Depth-image-based rendering(DIBR) is widely used in 3 DTV, free-viewpoint video, and interactive 3 D graphics applications. Typically, synthetic images generated by DIBR-based systems incorporate various distortions, particularly geometric distortions induced by object dis-occlusion. Ensuring the quality of synthetic images is critical to maintaining adequate system service. However, traditional 2 D image quality metrics are ineffective for evaluating synthetic images as they are not sensitive to geometric distortion. In this paper, we propose a novel no-reference image quality assessment method for synthetic images based on convolutional neural networks, introducing local image saliency as prediction weights. Due to the lack of existing training data, we construct a new DIBR synthetic image dataset as part of our contribution. Experiments were conducted on both the public benchmark IRCCyN/IVC DIBR image dataset and our own dataset. Results demonstrate that our proposed metric outperforms traditional 2 D image quality metrics and state-of-the-art DIBR-related metrics.
文摘SAR images commonly suffer fromspeckle noise,posing a significant challenge in their analysis and interpretation.Existing convolutional neural network(CNN)based despeckling methods have shown great performance in removing speckle noise.However,these CNN-basedmethods have a fewlimitations.They do not decouple complex background information in amulti-resolutionmanner.Moreover,they have deep network structures thatmay result in many parameters,limiting their applicability tomobile devices.Furthermore,extracting key speckle information in the presence of complex background is also a major problem with SAR.The proposed study addresses these limitations by introducing a lightweight pyramid and attention-based despeckling(PAN-Despeck)network.The primary objective is to enhance image quality and enable improved information interpretation,particularly on mobile devices and scenarios involving complex backgrounds.The PAN-Despeck network leverages domainspecific knowledge and integrates Gaussian Laplacian image pyramid decomposition for multi-resolution image analysis.By utilizing this approach,complex background information can be effectively decoupled,leading to enhanced despeckling performance.Furthermore,the attention mechanism selectively focuses on key speckle features and facilitates complex background removal.The network incorporates recursive and residual blocks to ensure computational efficiency and accelerate training speed,making it lightweight while maintaining high performance.Through comprehensive evaluations,it is demonstrated that PAN-Despeck outperforms existing image restoration methods.With an impressive average peak signal-to-noise ratio(PSNR)of 28.355114 and a remarkable structural similarity index(SSIM)of 0.905467,it demonstrates exceptional performance in effectively reducing speckle noise in SAR images.The source code for the PAN-DeSpeck network is available on GitHub.
基金This work is supported by the 13th Five-Year Plan for Jiangsu Education Science(D/2020/01/22)JSPIGKZ and Natural Science Research Projects of Colleges and Universities in Jiangsu Province(19KJB510022)。
文摘The traditional Range Doppler(RD)algorithm is unable to meet practical needs owing to the limit of resolution.The order of fractional Fourier Transform(FrFT)and the length of sampling signals affect SAR imaging performance when FrFT is applied to RD algorithm.To overcome the above shortcomings,the purpose of this paper is to propose a high-resolution SAR image algorithm by using the optimal order of FrFT and the sample length constraints for the range direction.The expression of the optimal order of SAR range signals via FrFT is deduced in detail.The initial sample length and its constraints are proposed to obtain the best sample length of SAR range signals.Experimental results demonstrate that,when the range sampling-length changes in a certain interval,the best sampling-length will be obtained,which the best values of the range resolution,PSLR and ISLR,will be derived respectively.Compared with traditional RD algorithm,the main-lobe width of the peak-point target of the proposed algorithm is narrow in the range direction.While the peak amplitude of the first side-lobe is reduced significantly,those of other side-lobes also drop in various degrees.
基金Support by the National Natural Science Foundation of China (NSFC) (No.60472072)the Specialized Research Foundation for the Doctoral Program of Higher Education (No.20040699034)+1 种基金the Aeronautical Science Foundation of China (No.05I53076)the Yellow River Conser-vancy Commission (YRCC) Research on ecological im-provement of the Yellow River (No.2004SZ01-04)
文摘There is difficulty for distinguishing of river and shadow in Synthetic Aperture Radar (SAR) images. A method of river segmentation in SAR images based on wavelet energy and gradient is proposed in this paper. It mainly includes two algorithms: coarse segmentation and refined segmen- tation. Firstly, The river regions are coarsely segmented by the wavelet energy feature,and then refined segmented accurately by the gradient threshold which is got adaptively. The experimental results show the validity of the method, which provides a good foundation for targets detection above the river.
