摘要
针对停车位置检测的问题,采用基于霍夫变换检测停车线,分割每个停车区域的图像预处理,使用VGG目标检测模型对每个停车区域进行迁移学习,判别相融合的空闲车位检测方法,对露天停车场高空定点摄像头传回的视频进行空闲车位识别与位置标注方面的研究。研究发现:VGG目标检测模型采用卷积神经网络减轻人工提取特征的工作量,与目标检测和识别中的经典机器学习方法相比,有较高的目标检测效率和准确度,为定点停车位的检测提供了一种实时位置反馈的解决方案。
To salve the problem of the parking space detection,the image preprocessing of each parking area for detect-ing parking line based on Hough transform is divided.The visual geometry group(VGG)target detection model is used to conduct transfer learning for each parking area to determine the car idle parking space detection.Research on the free parking space recognition and position marking is carried out for the high-altitude fixed-point camera in the open-air parking lot.It is found that the VGG target detection model in which CNN(convolutional neural networks)is used to simplify the workload of artificially extracting features,and has higher target detection efficiency and accuracy than the classical machine learning method in target detection and recognition.The study provides a real-time position feedback solution for the detection of fixed-point parking spaces.
作者
邢家源
张军
薛晨兴
雷雨婷
孙彦
XING Jia-yuan;ZHANG Jun;XUE Chen-xing;LEI Yu-ting;SUN Yan(School of Electronic Engineering,Tianjin University of Technology and Education,Tianjin 300222,China)
出处
《天津职业技术师范大学学报》
2019年第4期32-37,共6页
Journal of Tianjin University of Technology and Education
基金
天津市科技计划项目（14JCTPJC00537）
关键词
车位检测
图像预处理
迁移学习
VGG深度学习模型
卷积神经网络
parking space detection
image preprocessing
transfer learning
visual geometry group(VGG)deep learning model
convolutional neural network(CNN)
