基于深度学习的移动机器人目标自动跟随控制系统设计

Design of Automatic Target Following Control System for Mobile Robots Based on Deep Learning

摘要:

移动机器人在跟随运动目标时,容易受到周围环境的影响,导致目标识别准确性降低,从而影响自动跟随控制效果。为此,设计了基于深度学习的移动机器人目标自动跟随控制系统。系统框架设计为感知层、处理和控制层以及执行层。利用感知层中的视觉传感器、超声波传感器、MEMS传感器,采集信息并传输到处理和控制层,单片机处理器运行两个程序,前一个程序利用深度学习中的残差学习网络、深度卷积网络、长短期记忆神经网络进行图像处理和目标识别,后一个程序结合超声波传感器测距信息计算目标坐标。PLC微控制器承载控制程序,结合MEMS传感器采集到的角度信息,基于PID设计双环控制器,在其控制下实现移动机器人目标自动跟随控制。实验结果表明,无论在何种环境下,设计系统的误识别图像和未识别图像数量较少,角度跟随平均误差和跟随距离平均误差均较小,具有较好的目标识别功能和较强的抗环境干扰能力,更适用于多种环境的目标跟随,确保移动机器人目标自动跟随控制效果。

Abstract:

Mobile robots are easily affected by the surrounding environment when following moving targets, resulting in a decrease in target recognition accuracy and thus affecting the effectiveness of automatic following control. For this purpose, a deep learning based automatic target following control system for mobile robots was designed. The system framework is designed as a perception layer, processing and control layer, and execution layer. The visual sensor, ultrasonic sensor and MEMS sensor in the perception layer are used to collect information and transmit it to the processing and control layer. The MCU processor runs two programs. The former program uses the residual learning network, depth convolution network and Long short-term memory neural network in the depth learning to process images and identify targets. The latter program uses the ranging information of the ultrasonic sensor to calculate the target coordinates. The PLC microcontroller carries the control program, combines the angle information collected by MEMS sensors, and designs a dual loop controller based on PID to achieve automatic target following control of the mobile robot under its control. The experimental results show that, regardless of the environment, the number of misidentified and unrecognized images in the design system is relatively small, and the average error of angle following and following distance is relatively small. It has good target recognition function and strong ability to resist environmental interference, and is more suitable for target following in various environments, ensuring the effectiveness of automatic target following control for mobile robots.