4 个月前

改进深度学习方法在停车场占用检测中的应用

改进深度学习方法在停车场占用检测中的应用

摘要

近年来,作为智慧城市发展范式的一部分,停车引导系统已成为一种流行趋势。此类系统的关键部分是允许驾驶员在感兴趣的区域内搜索可用停车场的算法。传统的解决方法是将神经网络分类器应用于摄像头记录。然而,现有的系统在特定视觉条件下的泛化能力和适当测试方面表现不足。本研究对最先进的停车场占用检测算法进行了广泛评估,将其预测质量与最近出现的视觉变换器(Vision Transformers)进行了比较,并提出了一种基于EfficientNet架构的新流程。所进行的计算实验表明,我们的模型在性能上有所提升,并在5个不同的数据集上进行了评估。

代码仓库

eighonet/parking-research
官方
pytorch
GitHub 中提及

基准测试

基准方法指标
parking-space-occupancy-on-acmpsMobileNetV2
F1-score: 0.9971
parking-space-occupancy-on-acmpsCFEN
F1-score: 0.9789
parking-space-occupancy-on-acmpsResNet50
F1-score: 0.9379
parking-space-occupancy-on-acmpsCarNet
F1-score: 0.9877
parking-space-occupancy-on-acmpsEfficientNet-P
F1-score: 0.9982
parking-space-occupancy-on-action-cameraViT
F1: 0.8152
parking-space-occupancy-on-action-cameraVGG-19
F1-score: 0.9152
parking-space-occupancy-on-action-cameraMobileNetV2
F1-score: 0.9343
parking-space-occupancy-on-action-cameramAlexNet
F1-score: 0.8577
parking-space-occupancy-on-action-cameraResNet50
F1-score: 0.8377
parking-space-occupancy-on-action-cameraCFEN
F1-score: 0.8302
parking-space-occupancy-on-action-cameraEfficientNet-P
F1-score: 0.9125
parking-space-occupancy-on-cnrpark-extCFEN
F1-score: 0.8482
parking-space-occupancy-on-cnrpark-extEfficientNet-P
F1-score: 0.9683
parking-space-occupancy-on-cnrpark-extResNet50
F1-score: 0.938
parking-space-occupancy-on-cnrpark-extMobileNetV2
F1-score: 0.9663
parking-space-occupancy-on-cnrpark-extCarNet
F1-score: 0.9332
parking-space-occupancy-on-cnrpark-extViT
F1-score: 0.9176
parking-space-occupancy-on-cnrpark-extVGG-19
F1-score: 0.9629
parking-space-occupancy-on-pklotVGG-19
F1-score: 0.9988
parking-space-occupancy-on-pklotResNet50
F1-score: 0.9926
parking-space-occupancy-on-spklCarNet
F1-score: 0.7131
parking-space-occupancy-on-spklMobileNetV2
F1-score: 0.6937
parking-space-occupancy-on-spklCFEN
F1-score: 0.5367
parking-space-occupancy-on-spklViT
F1-score: 0.7335
parking-space-occupancy-on-spklEfficientNet-P
F1-score: 0.7393
parking-space-occupancy-on-spklVGG-19
F1-score: 0.6801
parking-space-occupancy-on-spklResNet50
F1-score: 0.6674

用 AI 构建 AI

从想法到上线——通过免费 AI 协同编程、开箱即用的环境和市场最优价格的 GPU 加速您的 AI 开发

AI 协同编程
即用型 GPU
最优价格
立即开始

Hyper Newsletters

订阅我们的最新资讯
我们会在北京时间 每周一的上午九点 向您的邮箱投递本周内的最新更新
邮件发送服务由 MailChimp 提供
改进深度学习方法在停车场占用检测中的应用 | 论文 | HyperAI超神经