摘要
幼儿守护机器人的目标检测技术具有广阔的应用检测空间,但传统监测技术多基于手工分类特征数据进行分析,难以保障数据的准确性。研究基于卷积神经网络提出机器人监测系统,充分发挥神经网络对目标图像的深层特征检测效果,并提出深度可分离卷积结构和Focal Loss损失函数改进后的单发多框检测器(Single Shot MultiBox Detector,SSD)目标检测模型,基于软硬件条件设计检测系统。实验结果表明,改进SSD模型的稳定损失值为0.56~2.98,时间损失情况小于SSD模型(5s<6.5 s),且其与原模型的小目标检测精度差值幅度达到了7.63%。改进SSD模型在三种场景下的平均识别率在95%以上,高于其他对比算法,且目标检测速度为48.29 ms,应用效果较好。该方法能极大程度上为幼儿守护机器人系统的实际应用拓展提供借鉴思路,极大程度为幼儿提供了一个安全有效的保障系统。
Target detection technology for child care robots has a broad application detection space,but traditional monitoring technologies are mostly based on manual classification feature data for analysis,which is difficult to ensure the accuracy of data.Research and propose a robot monitoring system based on convolutional neural networks to fully utilize the deep feature detection effect of neural networks on target images.Propose a single shot multi box detector(SSD) target detection model with a depth separable convolutional structure and an improved Focal Loss function,and design the detection system based on software and hardware conditions.The experimental results show that the stability loss value of the improved SSD model is 0.56-2.98,the time loss is smaller than that of the SSD model(5 s<6.5 s),and the difference in small target detection accuracy between the improved SSD model and the original model reaches 7.63%.The average recognition rate of the improved SSD model in three scenarios is above 95%,which is higher than other comparison algorithms.The target detection speed is 48.29 ms,and the application effect is good.This method can greatly provide reference ideas for the practical application expansion of the child care robot system,and greatly provide a safe and effective guarantee system for young children.
作者
贾丹妮
JIA Danni(Baoji Vocational and Technical College,Baoji city,Shaanxi 721013,China)
出处
《自动化与仪器仪表》
2023年第11期182-186,共5页
Automation & Instrumentation
基金
共青团陕西省委关于2022全省共青团和青年工作研究课题《“双减”背景下学校教育观念的演变态势研究——以宝鸡市为例》(20221802)。
