未来社区智慧泊车的少样本计数模型研究  

Research on Few-Shot Object Counting Model for Smart Parking in Future Communities

作　　者：张平 陈天来 陈照 王翼飞 ZHANG Ping;CHEN Tianlai;CHEN Zhao;WANG Yifei(China Railway Construction Electrification Bureau Group Co.Ltd.,Beijing 100043,China;Wuhan Haoguang Zhilian Technology Co.Ltd.,Wuhan Hubei 430074,China)

机构地区：[1]中国铁建电气化局集团有限公司,北京100043 [2]武汉昊广智联科技有限公司,湖北武汉430074

出　　处：《铁道建筑技术》2024年第1期10-12,25,共4页Railway Construction Technology

基　　金：中国铁建股份有限公司科技重大专项(2021-A04)。

摘　　要：在未来社区中如何更加人性化、智慧化地为居民提供泊车引导服务是未来社区建设的基础之一。本文提供了一种基于目标计数的模型来准确实时地检测空缺车位,为该服务提供技术支持,该模型采用基于卷积神经网络的密度图估计方法对车辆以及空车位进行计数,除为居民提供泊车引导外,还能够为社区管理提供汽车出入流量控制数据依据以及为宏观统筹端提供各社区数据以实现多社区车位共享;并对现有目标计数方法进行数据增强,试验结果表明模型在数据增强后MAE提升15.7%,RMSE提升25.81%。同时,该模型有较好的可扩展性,可以应用在其他领域。Ensuring a more user-friendly and intelligent provision of parking guidance services to residents is fundamental to future community development.This paper presents an object-counting-based approach for accurate and real-time detection of available parking spaces,offering technological support for this service.The model employs a convolutional-neural-network-based density estimation technique to count vehicles and available parking spaces.In addition to providing parking guidance for residents,it also serves as a source of traffic flow data for community management and facilitates macro-level data aggregation for multi-community parking space sharing;meanwhile,we perform data augmentation for the existing object counting method,and the experimental results show that the MAE of the model is improved by 15.7%and the RMSE is improved by 25.81%after data augmentation.Furthermore,this model demonstrats excellent versatility and can be applied to various other domains.

关 键 词：未来社区 目标计数 智慧泊车 少样本计数模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] U491.7[自动化与计算机技术—计算机科学与技术]