摘要
行人检测技术是众多领域的基础技术之一。例如无人驾驶汽车、智能视频监控和人体行为分析等领域。在智能视频监控领域中，行人检测对系统的实时性和准确性都有较高的要求，然而两者往往不能兼得，在计算能力一定的硬件环境下，高实时性意味着准确性会有所降低，高准确性往往需要降低实时性来获取。如何同时保证系统的实时性和准确性成为智能视频监控系统的一个难题。近年来，随着深度学习技术的不断突破，该问题得到了一定程度的解决。 深度学习技术作为人工智能的一个子领域，在计算机视觉中取得了很好的效果，基于深度学习的行人检测系统就是智能视频监控系统的一种。它是针对油井环境下的行人检测而设计和实现的系统。该系统将DarkNet框架作为训练时所使用的框架，将Yolo v2作为训练时所使用的网络模型。对客户提供的真实场景下的视频或图像数据进行标注和训练，将训练好的模型嵌入到四路智能分析设备TX1中，通过网络摄像头获取现场实时监控画面的视频数据，或者直接读取本地的视频数据，经过模型的处理得到视频画面中行人的位置信息，之后将视频信息以及检测结果传输到Web服务器，在网页端进行监控画面以及检测结果的显示。服务器以及深度学习框架都是在Linux系统下运行的，用户只需打开一个浏览器输入系统IP地址，登录系统即可进行相应的操作。
基于深度学习的行人检测系统能够对获取到的实时监控数据和本地视频数据进行检测，并且能够达到一定的实时性和准确性。用户操作方便，只需通过浏览器就可以整个系统进行操作。该系统可以在网页端实时显示视频和检测结果，可以控制系统是否进行检测，可以设置检测结果所发送的目的IP地址等。
关键词：深度学习Yolo v2网络结构DarkNet框架行人检测
Abstract
Pedestrian detection technology is one of the basic technologies in many fields.Such as unmanned vehicles,intelligent video surveillance and human behavior analysis.In the field of intelligent video surveillance,pedestrian detection in real-time and accuracy of the system have higher requirements,but they often have not,in the calculation of the ability of hardware environment,high real-time mean accuracy will be reduced,the high accuracy often need to reduce the real time to get.How to ensure the real-time and accuracy of the system at the same time has become a difficult problem in the intelligent video surveillance system.In recent years,with the continuous breakthrough of deep learning technology,this problem has been solved to a certain extent.
As a subfield of artificial intelligence,deep learning technology has achieved good results in computer vision.Pedestrian detection system based on deep learning is one of the intelligent video surveillance systems.It is a system designed and implemented for pedestrian detection under the oil well environment.The system uses the DarkNet framework as the framework used in training and uses the Yolo V2as a network model used in training.The real scene to provide the video or image data were tagged and training,embedding models trained to four intelligent analysis device in TX1,access to the video data monitoring picture through the network camera,or directly read the video data locally,through model processing by the position information of video images of pedestrians then,the video information and the detection result is transmitted to the Web server,display monitoring screen and test results in the end of the page.The server and deep learning framework are running under the Linux system.Users just need to open a browser to input the IP address of the system,and login the system to do the corresponding operation.
The pedestrian detection system based on deep learning can detect the real-time monitoring data and local video data,and achieve real-time and accuracy.The user is easy to operate,and the whole system can be operated only through the browser.The system
can display video and test results in real time at the webpage end,control whether the system detects and set the destination IP address sent by the test result.
Key words：Deep learning Yolo v2network structure DarkNet frame
Pedestrian detection
目录
摘要..............................................................................................................II 1绪论
1.1研究背景 (1)
摄像头测试1.2研究的目的和意义 (2)
1.3国内外研究概况 (3)
1.4主要工作 (4)
2关键技术介绍
2.1深度学习框架 (6)
2.2深度学习网络Yolo v2 (7)
2.3Gstreamer多媒体框架 (10)
2.4本章小结 (11)
3行人检测系统需求分析
3.1系统需求分析总体概述 (12)
3.2系统的功能性需求 (14)
3.3系统的非功能性需求 (16)
3.4本章小结 (17)
4行人检测系统的设计
4.1系统总体结构设计 (18)
4.2系统概要设计 (20)
4.3系统详细设计 (25)
4.4本章小结 (30)
5行人检测系统的实现与测试
5.1系统开发运行环境 (31)
5.2系统功能实现 (32)
5.3系统测试 (41)
5.4本章小结 (44)
6总结与展望
6.1全文总结 (45)
6.2展望 (46)
致谢 (47)
参考文献 (48)
豫公网安备41160202000603
站长QQ:729038198
关于我们
投诉建议