复杂场景中准确实时的人物识别算法研究
Research on Accurate and Real-time Character Recognition Algorithms in Complex Scenes
杨锦 1景飞 1张童童 1涂娅欣2
作者信息
- 1. 中国电子科技集团有限公司 第二十九研究所,四川 成都 610036;四川省宽带微波电路高密度集成工程研究中心,四川 成都 610036
- 2. 国网四川省电力公司 计量中心,四川 成都 610045
- 折叠
摘要
目前,基于深度学习的单步目标检测器已被广泛用于实时目标检测,但其对目标的定位精度较差,并且存在目标漏检、误检等问题.文章提出了一种用于复杂场景中准确实时的人物识别算法.首先,使用高斯YOLOv3 来估计预测框的坐标和定位不确定性,然后,采用基于注意力机制的非极大值抑制方法去除冗余的检测框,提高目标检测结果的准确性.经自建数据集训练并测试,改进的高斯YOLOv3 对人物的识别精度为 83.1%,比YOLOv3 提高了 1.68%,检测模型可以应用于军事战场人物的识别和定位,为战场态势感知系统提供有效的技术支持.
Abstract
Currently,single step object detectors based on Deep Learning have been widely used for real-time object detection,but their positioning accuracy for targets is poor,and there are problems such as missed detection and false detection of targets.This paper proposes an accurate and real-time character recognition algorithm for complex scenes.Firstly,this paper uses Gaussian YOLOv3 to estimate the coordinates and positioning uncertainty of the prediction box.Then,a Non-Maximum Suppression method based on Attention Mechanism is used to remove redundant detection boxes and improve the accuracy of target detection results.After self-built dataset training and testing,the improved Gaussian YOLOv3 has a character recognition accuracy of 83.1%,which is 1.68%higher than YOLOv3.The detection model can be applied to the recognition and positioning of military battlefield characters,providing effective technical support for battlefield situation awareness systems.
关键词
人物识别/高斯模型/注意力机制/高斯YOLOv3/非极大值抑制Key words
character recognition/Gaussian model/Attention Mechanism/Gaussian YOLOv3/Non-Maximum Suppression引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据