基于计算机视觉的摔倒检测系统的设计

发布时间：2018-01-07 12:07

本文关键词：基于计算机视觉的摔倒检测系统的设计　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：意外摔倒威胁着人们的健康生活,尤其是在人口老龄化问题日益突出的今天。为解决这一问题,研究者们提出基于可穿戴设备、环境传感器与计算机视觉这三种技术方案,其中计算机视觉技术逐渐成为室内检测环境下国际研究的主流方案。因此本文提出一种基于计算机视觉的摔倒检测算法,并通过编程设计出一款以摔倒检测、视频监控为主要功能的嵌入式摔倒检测系统。在摔倒检测算法中,通过单摄像头采集视频,混合高斯模型识别运动目标并从中提取特征变量以描述人的运动,最后对特征变量进行摔倒判决得到检测结果。在算法设计上本文从运动目标与场景环境这两方面提取动态特征与状态特征,并把动态特征与状态特征同阈值法与机器学习法这两类摔倒判决算法有效结合,发挥其各自在摔倒识别过程中的优越性。此外,本文讨论了对于物体遮挡下运动目标的摔倒判决处理并在图像预处理阶段对矩形框做出优化。系统采用WIFI网络构建客户端/服务器模式,客户端使用ARM+DSP双核异构嵌入式系统实现视频采集、摔倒检测、视频压缩、网络传输等功能;服务器使用个人计算机进行网络连接,摔倒报警与复位,视频解码播放与保存并以可视化界面进行操作。通过客户端与服务器间的数据传输,系统可实现摔倒检测算法对场景中发生的摔倒情况自动报警以及看护人员对网内各住户的远程监控。实验结果表明,摔倒检测算法在无遮挡条件下、透光物体遮挡下以及不同光照环境下的准确率分别可达94.2%、91.5%与91.0%,可见本设计中采用的摔倒检测算法能够在光照变化、透光物体遮挡等复杂环境下有效地检测出摔倒事件的发生,辅助以场景视频可使看护人员对摔倒者进行及时有效的救助。
[Abstract]:Accidental falls threaten people's healthy lives, especially at a time when the aging population is becoming more and more acute. To solve this problem, researchers have proposed a new approach based on wearable devices. Environmental sensor and computer vision are three technical solutions. Computer vision technology has gradually become the mainstream of international research in indoor detection environment. So this paper proposes a fall detection algorithm based on computer vision and designs a fall detection algorithm by programming. Video surveillance as the main function of the embedded fall detection system. Fall detection algorithm through a single camera to capture video mixed Gao Si model to identify moving targets and extract feature variables to describe human motion. Finally, the detection result is obtained by the fall decision of the feature variable. In the algorithm design, the dynamic feature and the state feature are extracted from the moving object and scene environment. The dynamic feature and the state feature are combined with the threshold method and the machine learning method to effectively combine the two kinds of fall decision algorithms to give play to their respective advantages in the process of fall recognition. This paper discusses the fall decision processing of moving object under object occlusion and optimizes the rectangle frame in the image preprocessing stage. The system uses WIFI network to construct client / server mode. The client uses ARM DSP dual-core heterogeneous embedded system to realize the functions of video acquisition, fall detection, video compression, network transmission and so on. The server uses a personal computer for network connection, a fall alarm and a reset, video decoding, playback, and operation with a visual interface. Data transmission between the client and the server is carried out. The system can realize the automatic alarm of the fall in the scene and the remote monitoring of the residents in the network by the fall detection algorithm. The experimental results show that the fall detection algorithm is in the condition of no occlusion. The accuracy of transparent object occlusion and different light environment can reach 91.5% and 91.0% respectively. It shows that the fall detection algorithm used in this design can change in illumination. In the complex environment such as transparent object occlusion, the fall events can be detected effectively, and the scene video can help the caregivers to rescue the fall in a timely and effective manner.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】