目的在智能监控视频分析领域中,行人重识别是跨无交叠视域的摄像头匹配行人的基础问题。在可见光图像的单模态匹配问题上,现有方法在公开标准数据集上已取得优良的性能。然而,在跨正常光照与低照度场景进行行人重识别的时候,使用可见光...目的在智能监控视频分析领域中,行人重识别是跨无交叠视域的摄像头匹配行人的基础问题。在可见光图像的单模态匹配问题上,现有方法在公开标准数据集上已取得优良的性能。然而,在跨正常光照与低照度场景进行行人重识别的时候,使用可见光图像和红外图像进行跨模态匹配的效果仍不理想。研究的难点主要有两方面:1)在不同光谱范围成像的可见光图像与红外图像之间显著的视觉差异导致模态鸿沟难以消除;2)人工难以分辨跨模态图像的行人身份导致标注数据缺乏。针对以上两个问题,本文研究如何利用易于获得的有标注可见光图像辅助数据进行单模态自监督信息的挖掘,从而提供先验知识引导跨模态匹配模型的学习。方法提出一种随机单通道掩膜的数据增强方法,对输入可见光图像的3个通道使用掩膜随机保留单通道的信息,使模型关注提取对光谱范围不敏感的特征。提出一种基于三通道与单通道双模型互学习的预训练与微调方法,利用三通道数据与单通道数据之间的关系挖掘与迁移鲁棒的跨光谱自监督信息,提高跨模态匹配模型的匹配能力。结果跨模态行人重识别的实验在“可见光—红外”多模态行人数据集SYSU-MM01(Sun Yat-Sen University Multiple Modality 01)、RGBNT201(RGB,near infrared,thermal infrared,201)和RegDB上进行。实验结果表明,本文方法在这3个数据集上都达到领先水平。与对比方法中的最优结果相比,在RGBNT201数据集上的平均精度均值mAP(mean average precision)有最高接近5%的提升。结论提出的单模态跨光谱自监督信息挖掘方法,利用单模态可见光图像辅助数据挖掘对光谱范围变化不敏感的自监督信息,引导单模态预训练与多模态有监督微调,提高跨模态行人重识别的性能。展开更多
The recently invented artificial bee colony （ABC） al- gorithm is an optimization algorithm based on swarm intelligence that has been used to solve many kinds of numerical function optimization problems. It performs ...The recently invented artificial bee colony （ABC） al- gorithm is an optimization algorithm based on swarm intelligence that has been used to solve many kinds of numerical function optimization problems. It performs well in most cases, however, there still exists an insufficiency in the ABC algorithm that ignores the fitness of related pairs of individuals in the mechanism of find- ing a neighboring food source. This paper presents an improved ABC algorithm with mutual learning （MutualABC） that adjusts the produced candidate food source with the higher fitness between two individuals selected by a mutual learning factor. The perfor- mance of the improved MutualABC algorithm is tested on a set of benchmark functions and compared with the basic ABC algo- rithm and some classical versions of improved ABC algorithms. The experimental results show that the MutualABC algorithm with appropriate parameters outperforms other ABC algorithms in most experiments.展开更多
文摘目的在智能监控视频分析领域中,行人重识别是跨无交叠视域的摄像头匹配行人的基础问题。在可见光图像的单模态匹配问题上,现有方法在公开标准数据集上已取得优良的性能。然而,在跨正常光照与低照度场景进行行人重识别的时候,使用可见光图像和红外图像进行跨模态匹配的效果仍不理想。研究的难点主要有两方面:1)在不同光谱范围成像的可见光图像与红外图像之间显著的视觉差异导致模态鸿沟难以消除;2)人工难以分辨跨模态图像的行人身份导致标注数据缺乏。针对以上两个问题,本文研究如何利用易于获得的有标注可见光图像辅助数据进行单模态自监督信息的挖掘,从而提供先验知识引导跨模态匹配模型的学习。方法提出一种随机单通道掩膜的数据增强方法,对输入可见光图像的3个通道使用掩膜随机保留单通道的信息,使模型关注提取对光谱范围不敏感的特征。提出一种基于三通道与单通道双模型互学习的预训练与微调方法,利用三通道数据与单通道数据之间的关系挖掘与迁移鲁棒的跨光谱自监督信息,提高跨模态匹配模型的匹配能力。结果跨模态行人重识别的实验在“可见光—红外”多模态行人数据集SYSU-MM01(Sun Yat-Sen University Multiple Modality 01)、RGBNT201(RGB,near infrared,thermal infrared,201)和RegDB上进行。实验结果表明,本文方法在这3个数据集上都达到领先水平。与对比方法中的最优结果相比,在RGBNT201数据集上的平均精度均值mAP(mean average precision)有最高接近5%的提升。结论提出的单模态跨光谱自监督信息挖掘方法,利用单模态可见光图像辅助数据挖掘对光谱范围变化不敏感的自监督信息,引导单模态预训练与多模态有监督微调,提高跨模态行人重识别的性能。
基金supported by the National Natural Science Foundation of China (60803074)the Fundamental Research Funds for the Central Universities (DUT10JR06)
文摘The recently invented artificial bee colony （ABC） al- gorithm is an optimization algorithm based on swarm intelligence that has been used to solve many kinds of numerical function optimization problems. It performs well in most cases, however, there still exists an insufficiency in the ABC algorithm that ignores the fitness of related pairs of individuals in the mechanism of find- ing a neighboring food source. This paper presents an improved ABC algorithm with mutual learning （MutualABC） that adjusts the produced candidate food source with the higher fitness between two individuals selected by a mutual learning factor. The perfor- mance of the improved MutualABC algorithm is tested on a set of benchmark functions and compared with the basic ABC algo- rithm and some classical versions of improved ABC algorithms. The experimental results show that the MutualABC algorithm with appropriate parameters outperforms other ABC algorithms in most experiments.
