Deep-learning SSD-based Smart Touch Interface and Interactive Robot Design

• Main Authors: Chen, Hsu-Huan, 陳緒桓
• Other Authors: Cheng, Chih-Yung
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/cbxmkw

碩士 === 國立臺灣海洋大學 === 電機工程學系 === 107 === The main purpose of this thesis is to study the application of the deep learning object recognition to the control of several devices including omnidirectional mobile robot, delta robot arm and household appliances. The SSD (Single Shot multibox Detector) neural network is used to train and identify different objects to be controlled and then the corresponding manipulations can be performed. The system architecture can mainly be divided into a main control console and several controlled devices. The main control module uses the tablet computer combined with the neural network model to process the images captured through the front lens. Multiple target objects appearing in the input image can be identified for the category and position respectively. Furthermore, user can connect and control his/her interested identified item on the screen with Bluetooth interface. The three items on the current list of controlled devices all use Arduino pro mini as the control core and HC-05 Bluetooth to communicate with console. Omnidirectional mobile robot and delta robot arm both use three servo motors. The former uses servo motors to drive car and the latter to control the robot arm operations. When the tablet computer is connected with omnidirectional mobile robot, user can drag it to move in all directions with one finger or rotate body posture with two fingers on the screen. If system is connected with delta robot, user can click on the screen for cat or dog pictures placed under the robot arm and robot arm will perform classification and collection. Finally, if connected with desk lamp, user can click the lamp on the screen to control it open or close