Design of target positioning system for monocular vision manipulator based on neural network

Abstract

a target space positioning system based on target detection for monocular camera manipulator was developed in this research. The system uses EdgeBoard embedded AI computing card as the core controller, adopts Linux operating system, and using Paddle-Lite architecture under Linux system to predict the deployment of target detection neural network. The control core of the drive part is ESP32, using OV2640 camera module and PCA9685 as system input and output modules. Communication between upper and lower computers adopts UDP and TCP network communication protocols. Target recognition uses mobilenet-ssd neural network based on paddle framework. The robotic arm system calculates the posture of the robotic arm by establishing a spatial D-H coordinate system. According to the position of the target output by the neural network, inversely calculate the spatial position of target through multiple spatial positioning of the robotic arm. It can realize the task of grabbing and categorizing the target.