基于卷积神经网络的非图像的手势识别控制系统设计 毕业论文+源码及原理图+流程图及源文件+用户手册+设计说明文档+演示视频+论文检测查重报告
摘 要
本文设计并实现了一个基于卷积神经网络的非图像手势识别控制系统，该系统利用MPU6050陀螺仪实时采集时间序列的加速度传感器数据，通过一维卷积神经网络（1D CNNs）对用户手势动作进行识别。该系统旨在解决传统图像手势识别方法在低光照、复杂背景等环境下识别不准确的问题。
设计采用了STM32F103CBT6作为主控芯片，实现了手势数据的采集、处理与识别，以及基于识别结果对外部设备的控制。系统集成了射频遥控和蓝牙通信模块，能够识别包含10种手势的手势库，如特定字母和数字，且正确识别率控制在90%以上。此外，系统还具备高度的可扩展性，用户可以根据需要添加新的手势或功能模块。本研究不仅提高了手势识别的准确性和适应性，也为非图像手势识别技术的应用提供了新的思路和方法。
关键词：卷积神经网络（CNN）；非图像手势识别；时间序列数据；MPU6050陀螺仪；STM32
Abstract
In this paper, a non-image gesture recognition control system based on convolutional neural network is designed and implemented. This system uses MPU6050 gyroscope to collect time series acceleration sensor data and identify the user gesture action through one-dimensional convolutional neural network (1D CNNs). The system aims to solve the problem of inaccurate recognition of traditional image gesture recognition methods in low illumination and complex background environment.
The design uses STM32F103CBT6 as the master chip to realize the acquisition, processing and recognition of gesture data, and the control of external devices based on the recognition results. The system integrates the RF remote control and Bluetooth communication module, which can recognize the gesture library containing 10 gestures, such as specific letters and numbers, and the correct recognition rate is controlled at more than 90%. In addition, the system is highly scalable, allowing users to add new gestures or function modules as needed. This study not only improves the accuracy and adaptability of gesture recognition, but also provides new ideas and methods for the application of non-image gesture recognition technology.
Key words: convolutional neural network (CNN); non-image gesture recognition; time series data; MPU6050 gyroscope; STM 32
目 录
