基于多特征融合的人体行为识别技术研究

发布时间：2018-08-02 20:27

【摘要】：随着MEMS技术的不断发展,基于传感器的人体行为识别技术得到了飞速发展,在诸多领域得到了广泛应用,如运动检测、康复医疗、能耗评估等。相比于基于计算机视觉的人体行为识别,基于传感器的方法更能体现运动本质,在便携性、舒适性、成本及易维护性方面优势突出。虽然基于传感器的人体行为识别在各方面都取得了长足发展,但仍有一些关键问题需要解决,如怎样提取更加有效的特征,如何构建高精度、低复杂度的分类器等。围绕这些问题,本文主要做了以下工作:1、了解了人体行为识别领域和步态相位识别领域国内外的研究现状,总结了人体行为识别方法和步态相位识别方法,对各方法的优势和不足进行了分析,并根据研究目标确定自己的研究方法。2、设计了一套由加速度和陀螺仪数据采集装置及Pedar-X分布式足底压力鞋垫两部分组成的传感器数据采集系统用于数据采集。其中,加速度和陀螺仪数据采集装置由STM32单片机、MPU6050陀螺仪和加速度计和nRF24L01无线通讯模块组成,可以实时采集人体的加速度及角速度数据,并上传到上位机;Pedar-X分布式足底压力鞋垫可实现足底压力数据的实时采集功能。3、在步态相位识别方面,针对复杂路况步态相位识别的需求,提出了一种基于多特征融合的复杂路况步态识别方法。以复杂路况的步态相位识别为研究对象,融合足底平均压力、足底压力中心点和大腿角度特征,用有向无环图支持向量机的方法进行分析,有效识别了支撑前期、支撑中期、支撑后期、摆动前期、摆动后期五个相位。4、在人体行为识别方面,为了更好的表征人体的行为,提出了一种基于多特征融合的人体行为识别方法。该方法融合加速度特征和足底压力特征,用分层支持向量机方法识别了站立、坐下、行走、上楼、下楼和跑步六种行为,总体识别正确率达到92%以上。
[Abstract]:With the development of MEMS technology, the technology of human behavior recognition based on sensor has been developed rapidly, and has been widely used in many fields, such as motion detection, rehabilitation medical treatment, energy consumption evaluation and so on. Compared with the recognition of human behavior based on computer vision, the sensor-based approach can reflect the nature of motion, and has the advantages of portability, comfort, cost and maintainability. Although human behavior recognition based on sensors has made great progress in all aspects, there are still some key problems to be solved, such as how to extract more effective features, how to construct high-precision and low-complexity classifiers and so on. Around these problems, this paper mainly do the following work: 1, understand the research status in the field of human behavior recognition and gait phase recognition at home and abroad, summarize the methods of human behavior recognition and gait phase recognition. The advantages and disadvantages of each method are analyzed. According to the research objective, a sensor data acquisition system composed of acceleration and gyroscope data acquisition device and Pedar-X distributed sole pressure insole is designed for data acquisition. Among them, the acceleration and gyroscope data acquisition device is composed of STM32 single chip microcomputer MPU6050 gyroscope, accelerometer and nRF24L01 wireless communication module, which can collect the acceleration and angular velocity data of human body in real time. And uploading to the upper computer Pedar-X distributed plantar pressure insole can realize the real-time acquisition function of plantar pressure data. In the aspect of gait phase recognition, it aims at the need of gait phase recognition in complex road conditions. A gait recognition method for complex road conditions based on multi-feature fusion is proposed. Based on the gait phase recognition of complex road conditions, the mean plantar pressure, the center of plantar pressure and the angle of thigh are fused, and the method of directed acyclic graph support vector machine is used to effectively identify the prophase and middle stage of support. In the aspect of human behavior recognition, in order to better represent human behavior, a human behavior recognition method based on multi-feature fusion is proposed. This method combines acceleration characteristics and plantar pressure features, and uses hierarchical support vector machine to identify six behaviors: standing, sitting, walking, going upstairs, going downstairs and running. The overall recognition accuracy is over 92%.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212.9

【相似文献】