一种基于多光谱特征自适应融合的行人检测算法

马凯; 李洋; 周飞飞

一种基于多光谱特征自适应融合的行人检测算法

投稿时间：2024-11-05 修订日期：2025-12-01 点此下载全文

引用本文：

摘要点击次数: 15

全文下载次数: 0

作者	单位	邮编
马凯	南京南瑞信息通信科技有限公司	211100
李洋^*	南京南瑞信息通信科技有限公司
周飞飞	南京南瑞信息通信科技有限公司

中文摘要:摘要：多光谱数据因其能整合不同图像互补信息，已被广泛应用于自动驾驶、安全防卫等多个重要领域。其中，多光谱行人检测技术作为关键的组成部分，旨通过有效利用可见光与红外图像信息，提升复杂多变环境下的行人检测准确性。尽管近年来，多光谱行人检测技术取得了较为显著的进步，但仍面临诸多挑战。首先，多光谱数据集的稀缺限制了模型训练的泛化能力；其次，由于红外图像成像原理这一特性导致了图像对比度低和背景辐射强检测难度增大；最后，在实际应用中多光谱特征融合方法需要同时处理多个输入，进一步造成了模型特征冗余和检测速度降低的缺陷。针对上述问题，本文通过研究红外图像特征增强和多光谱行人特征融合方法，提出了一种渐进式特征红外图像增强处理方法，增强了红外行人目标，再对增强后的特征进行自适应融合，提高模型对行人目标的表达能力。实验结果表明，该算法对比前沿的检测算法在检测速度和检测精度上都有优势。

中文关键词:多光谱行人检测图像增强特征增强特征融合

A pedestrian detection algorithm based on adaptive fusion of multispectral features

Abstract:Abstract：Multispectral data has been widely used in fields such as autonomous driving and safety defense due to its ability to integrate complementary information from different images. Among them, multispectral pedestrian detection technology is a key component aimed at effectively utilizing visible light and infrared image information to improve pedestrian detection performance in complex environments. Although significant progress has been made in multispectral pedestrian detection technology in recent years, it still faces many challenges. Firstly, the scarcity of multispectral datasets limits the generalization ability of models. Secondly, the low image contrast and strong background radiation detection difficulty of infrared imaging principles increase. Finally, in practical applications, multispectral feature fusion methods need to handle multiple inputs, which can lead to feature redundancy and slow detection speed. In response to the above issues, this article studies feature enhancement methods for infrared images and multispectral pedestrian feature fusion methods. A progressive feature based infrared image enhancement method is proposed to enhance infrared pedestrian targets. The enhanced features are adaptively fused to improve the expression ability of pedestrian targets and reduce redundant features. The experimental results show that the algorithm has advantages in detection speed and detection accuracy compared to state-of-the-art detection algorithms.

keywords:multispectral pedestrian detection image enhancement Feature enhancement Feature fusion

查看全文 查看/发表评论 下载pdf阅读器