摘要
针对户外停车场场景复杂,导致车位使用情况检测效能低、时效差和成本高的问题,提出了基于深度学习的车位实时检测方法。在计算能力有限的嵌入式系统中,对AlexNet网络部分进行修改,使其能直接在嵌入式系统中运行;在采集和截取到每个车位后,嵌入式系统对车位分类得出使用情况。实验表明,所提出的车位使用情况检测方法与传统方法相比较,精确度高,泛化能力强,在不同天气场景中也有很好的性能,并具有功耗低、易实现和成本低等特点。
In view of problems such as low efficiency,poor timeliness and high costs of detecting parking spaces caused by complex scenarios at parking lots,a solution for real-time detection of parking spaces was presented based on deep learning in this paper.In an embedded system with limited computing resources,the AlexNet was partly modified for direct operation.After acquiring and intercepting each parking space,the embedded system classified parking spaces to obtain the usage situation.The experiment shows that the method for detecting the usage of parking spaces presented in this paper is highly accurate and capable of being popularized,performing very well in different weather conditions,and boasting such features as low power consumption,easy implementation and low costs,compared with traditional methods.
作者
龙劲峄
周骅
LONG Jinyi;ZHOU Hua(College of Big Data and Information Engineering, Guizhou University, Guiyang 550025, China)
出处
《中国科技论文》
CAS
北大核心
2021年第3期295-300,共6页
China Sciencepaper
基金
贵州大学引进人才科研项目(贵大人基合字(2015)53号)
贵州大学培育项目(黔科合平台人才[2017]5788-60)。
关键词
图像处理
深度学习
卷积神经网络
停车场
嵌入式系统
车位
image processing
deep learning
convolutional neural networks(CNN)
parking lot
embedded system
parking space