军事知识服务平台
面向复杂遮挡环境的双目单光子人体检测算法
Binocular Single-Photon Human Detection Algorithm for Complex Occlusion Environments
摘要 面向城市反恐等高风险场景,亟需在非接触条件下对窗后人体目标实现快速、准确的姿态识别与精确定位,而现有依赖主动激光照射成像后进行图像识别的透窗系统难以处理复杂遮挡下的低信噪比回波数据.为此,基于双目单光子激光透窗系统,提出一种适用于目标回波分类的一维时序直方图的R-UNet网络结构.首先,在前端基于目标回波的噪声特性使用轻量化的残差网络自适应地抑制背景噪声,同时保留高频信号细节.随后,在骨干部分设计两级UNet结构捕捉回波信号中的多尺度时间依赖关系,确保微弱的目标信号特征得到增强.最后,通过末端的全连接头将高层特征映射至标签空间完成分类任务.为验证所提方法的有效性,构建了一个采集自真实场景的综合数据集,包含9种不同的目标类别.实验结果表明,所设计的网络相较于基线网络在保持了超过99%的分类准确率的同时,减少了65.1%的推理时间以及91.59%的参数量,更适合在边缘端进行部署.
Abstract:
In high-risk scenarios,such as urban counterterrorism,rapid and accurate pose recognition and precise localization of human targets behind windows under noncontact conditions are urgently required.However,existing through-window systems,which typically rely on post-imaging recognition algorithms following active laser illumination,struggle to process low signal-to-noise ratio echo data under complex occlusion conditions.To overcome this challenge,we propose an R-UNet architecture designed for target echo classification using one-dimensional temporal histograms,implemented in a binocular single-photon laser through-window system.Firstly,a lightweight residual network was employed at the front end to adaptively suppress background noise based on the noise characteristics of the target echoes while preserving high-frequency signal details.Subsequently,a two-stage UNet structure was designed within the backbone to capture multiscale temporal dependencies in the echo signals,thereby enhancing weak target signal features.Finally,a fully connected head mapped high-level features onto the label space to complete the classification task.To validate the effectiveness of the proposed method,we constructed a comprehensive dataset comprising 9 distinct target categories collected from real-world scenarios.Experimental results demonstrate that the proposed network achieves a classification accuracy greater than 99%.Furthermore,compared with baseline networks,this network reduced the inference time by 65.1%and the parameter count by 91.59%,making it highly suitable for edge deployment.
刊 名 半导体光电 ISTICPKU
年,卷(期): 2026, 47(5)
分类号: TP391.41
机标分类号: TP391.41 TN967.1 TP242.62
在线出版日期: 2026年11月4日
基金项目: 24-ZZCX-JDZ-11:国防科技大学自主科研基金项目,