Human-computer interaction gesture control recognition for high-performance embedded ai computing platform
面向高性能嵌入式人工智能计算平台的人机交互手势控制识别

摘要

本文主要介绍一种基于嵌入式AI计算平台的手势控制识别方法，利用红外摄像头和AI处理单元对手势进行识别和跟踪。使用红外摄像头作为检测工具可以实现低成本、易用等特点。该算法首先采用基于颜色模型的手势识别方法对输入的手势图像进行颜色转换并提取相应的肤色特征；然后通过对肤色特征进行加权，获得更丰富的颜色信息；最后通过计算颜色特征与肤色特征的相似度实现手势的识别。本文设计了一种基于红外摄像头和AI处理单元的手势控制识别系统。本文采用基于深度学习的目标检测算法，利用卷积神经网络（CNN）对目标进行检测，在不增加硬件成本的情况下实现手势识别与跟踪。在训练过程中，利用CNN提取手势图像的颜色特征和颜色信息进行训练；在测试过程中，使用相同数量和尺寸的红外摄像头对训练好的模型进行测试。此外，为了实现低成本、易用的特点，本文采用了基于深度学习的目标检测算法来验证手势识别的结果。

Abstract

This article mainly introduces a gesture control recognition method based on embedded AI computing platform, which uses infrared cameras and AI processing units to recognize and track gestures. Using infrared cameras as detection tools can achieve low-cost, easy-to-use and other characteristics. In this algorithm, firstly, the Gesture recognition method based on color model is used to convert the color of the input gesture image and extract the corresponding skin color features; Then, by weighting the skin color features, richer color information is obtained; Finally, the recognition of gestures was achieved by calculating the similarity between color features and skin tone features. This article designs a gesture control recognition system based on infrared cameras and AI processing units. This article adopts a deep learning based object detection algorithm. The algorithm uses Convolutional Neural Network (CNN) to detect the target, and realizes Gesture recognition and tracking without increasing the hardware cost. In the training process, CNN is used to extract the color features and color information of gesture images for training, while in the testing process, infrared cameras of the same number and size are used to test the trained model. In addition, in order to achieve the characteristics of low cost and ease of use, this paper uses the target detection algorithm based on deep learning to verify the results of Gesture recognition.