基于深度学习的机器人目标识别实验教学设计与改革

Teaching design and reform of robot target recognition experiment based on deep learning

设计了一个自追踪摄像头云台控制机器人实验。文章从人体的检测定位及云台控制方面阐述了系统的设计思路,利用机器视觉技术得到了人体目标在视角中的坐标,又对舵机进行了目标控制。追踪目标的过程最终以Mobile Net-SSD网络模型为基础,通过深度可分离的整合设计,改善了现有整合网络的复杂参数问题,从而优化了网络结构。后期将引导学生从改进本实验模型结构和图像预处理等方面提高最终算法的检测速度、检测准确率、识别准确率等。该实验设计涉及学科较为综合,功能扩展性好,实用性强,是机器人教学、机器学习实践与嵌入式系统相融合的创新实验项目。

A self tracking camera pan tilt control robot experiment is designed. This paper expounds the design idea of the system from the aspects of human body detection and positioning and pan tilt control, obtains the coordinates of the human body target in the perspective by using machine vision technology, and controls the target of the steering gear. The process of tracking the target is finally based on the mobilenet SSD network model.Through the deep separable integration design, the complex parameter problem of the existing integration network is improved, so as to optimize the network structure. Later, students will be guided to improve the detection speed,detection accuracy and recognition accuracy of the final algorithm from the aspects of improving the structure of the experimental model and image preprocessing. The experimental design involves relatively comprehensive disciplines, with good functional expansibility and high practicability. It is an innovative experimental project integrating robot teaching, machine learning practice and embedded system.