基于深度学习的四旋翼无人机控制系统设计

Design of Quadrotor UAV Control System Based on Deep Learning

摘要:

针对传统四旋翼无人机控制系统受到外界干扰，无法及时躲避障碍物而导致控制精准度低的问题，提出了基于深度学习的四旋翼无人机控制系统设计。根据四旋翼无人机控制系统总体结构，加入超声波测距模块。依据系统硬件框图，采用TMS320F28335型号主控芯片，实现关键态势智能分析。以串级 PID 控制器的控制对象为无人机姿态角度，控制电机转速。根据DSP发出不同占空比的PWM信号，改变无人机飞行姿态，依据执行机构驱动原理，保证无人机飞行时的平衡状态。使用红外遥控系统，应用编/解码操控集成电路芯片，采用TS0P1738型号红外线接收器，适合于红外线遥控数据传输。构建深度学习目标控制模型，利用处突阵法与三角形相似原理，计算像素尺寸，获取障碍物距无人机当前位置距离，避免受到外界障碍物干扰。自适应扩展Kalman滤波器技术对无人机自动控制系统有效减小测量误差，准确地对机动目标进行追踪。由系统调试结果可知，该系统控制的俯仰角、航向角、横滚角与实际值一致，对处理突发性群体事件具有重要意义。

Abstract:

In order to solve the problem that the control system of traditional four rotor UAV is disturbed by the outside world and can't avoid the obstacles in time, which leads to the low control accuracy, the control system design of four rotor UAV Based on deep learning is proposed. According to the overall structure of the four rotor UAV control system, the ultrasonic ranging module is added. According to the system hardware block diagram, TMS320F28335 main control chip is used to realize the key situation intelligent analysis. Taking the control object of the cascade PID controller as the attitude angle of the UAV, the motor speed is controlled. According to the PWM signals of different duty cycle sent by DSP, the flight attitude of UAV is changed, and according to the driving principle of actuator, the balance state of UAV during flight is ensured. Using infrared remote control system, using encoder / decoder to control IC chip, using ts0p1738 infrared receiver, suitable for infrared remote control data transmission. The depth learning target control model is constructed, and the pixel size is calculated by using the method of location matrix and the principle of triangle similarity. The distance between the obstacle and the current position of UAV is obtained to avoid the interference of external obstacles. The adaptive extended Kalman filter technology can effectively reduce the measurement error of UAV automatic control system and accurately track the maneuvering target. According to the results of system debugging, the pitch angle, heading angle and roll angle controlled by the system are consistent with the actual values, which is of great significance to deal with unexpected group events.