摘要
行人检测是计算机视觉领域重要的研究方向之一。随着深度学习的快速发展,基于可见光图像的行人检测技术的性能有了明显的提升,但在夜间或恶劣天气条件下检测性能急剧下降。基于可见光和热红外图像融合的行人检测可以有效解决全时段行人检测的问题,而多光谱行人检测的性能取决于可见光和热红外两种模态融合的机制。本文以基于深度学习的多光谱行人检测技术为研究对象,首先介绍了多光谱行人检测的基本深度学习模型,其次从特征级融合、决策级融合和模态迁移融合三个方面具体分析了深度学习在多光谱行人检测中的研究现状,最后分析了深度学习在多光谱行人检测中的应用前景。
Pedestrian detection is one of the important research directions in thefield of computer vision.With the rapid development of deep learning,the performance of pedestrian detection techniques based on visible images has improved significantly,but the detection performance decreases sharply at night or under bad weather conditions.Pedestrian detection based on visible and thermal infrared image fusion can effectively solve the problem of full time pedestrian detection,while the performance of multispectral pedestrian detection depends on the mechanism of fusion of both visible and thermal infrared modalities.In this paper,we take deep learning-based multispectral pedestrian detection technology as the research object,firstly,we introduce the basic deep learning model for multispectral pedestrian detection,secondly,we specifically analyze the research status of deep learning in multispectral pedestrian detection from three aspects:feature-level fusion,decision-level fusion and modal migration fusion,and finally,we analyze the application prospects of deep learning in multispectral pedestrian detection.
作者
秦君
李晓敏
夺实祥伟
Qin Jun;Li Xiaomin;Duo Shixiangwei(Dehong Power Supply Bureau of Yunnan Power Grid Co.,Ltd.,Dehong 678400,China)
出处
《云南电力技术》
2023年第6期54-60,共7页
Yunnan Electric Power
关键词
多光谱行人检测
深度学习
全时段检测
multispectral pedestrian detection
deep learning
full time detection
