基于深度学习的智能停车车位检测

Deep Learning Based Packing Space Detection in Intelligent Parking System

智能停车系统中车位检测易受天气变化、障碍物遮挡和背景干扰等影响而导致车位检测准确率降低,为了解决该问题,文中研究了基于深度学习的停车场车位占泊检测技术。首先,给出了基于改进YOLO的单车位检测算法,其把YOLO网络每层的输入归一化处理以有效避免过拟合,且把全连接层替换为全卷积层以提取多种尺寸输入的特征;然后,研究了基于改进SSD的多车位检测算法,其采用优化的ResNet网络替换原SSD的VGG网络为且优化了激活函数。实验和测试结果表明,所提方法能够有效提高单/多车位检测准确率。

To solve the problem of parking space detection in intelligent parking management system under the environment of weather change,obstacle occlusion and complex background interference,the deep learning based parking space occupation detection technology is investigated in this paper.Firstly,we study an improved YOLO single parking space detection algorithm,which normalizes the input of each layer of the network to effectively avoid over fitting,and replaces the YOLO full connection layer with the full convolution layer to extract the features of various input sizes.Then,we explore the multi-parking space detection algorithm based on the improved SSD,which replaces the VGG network of the original SSD with the optimized ResNet network and applies optimized activation functions.The experimental results show that the proposed method can effectively improve the accuracy of parking space detection.