Development of an autonomous camera-gimbal system for target following or artifact detection applications

• 主要作者: Gnanaprakasam Naveen
• 其他作者: Xie Lihua
• 格式: Thesis-Master by Coursework
• 語言: English
• 出版: Nanyang Technological University 2020
• 主題: Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics
• 在線閱讀: https://hdl.handle.net/10356/142947

Object detection is a fundamental requirement for a robot to perform tasks in target following applications. With increase in human-robot collaboration in various fields, the usage of robotic technologies in numerous settings and situations such as mapping and inspection, has widely increased. The major challenge in building a completely autonomous system is to enable the on-board camera, which is usually attached to a gimbal device to explore and detect targets in the environment. Cameras are one of the key components in object detection, so it is very important to build an efficient camera gimbal system to detect targets or objects in different environments and conditions. The objective of this research work is to design a camera-gimbal system and implement image processing techniques and deep learning pipeline to autonomously detect targets or objects, depending on the application. On successful development, the camera gimbal system has been tested in different scenarios to measure the efficiency of object detection in the developed system. Ten different models were tested in the developed system and suitable models were selected based on the speed and accuracy. RFCN Resnet101 and Faster RCNN Resenet101 models were selected for their consistent speed of 30-32 milli seconds and accuracy measure of 96-99%. Various tests were conducted on the camera system, with the selected deep learning models, to validate the reliability and robustness of the developed system. As a result, the developed system was proven to be potentially placed both on ground and aerial mobile robots. This research work focuses on improvement of accuracy of autonomous object detection by utilizing the camera gimbal system under different scenarios to help in increasing the performance of target following and detection applications.