İHA Üzerinden Derin Öğrenme Tabanlı Nesne Belirlenmesi ve GNSS Konum Koordinatlarının Hesaplanması

Erdal H. (Executive) , Doğan B. , Tiryaki F.

Project Supported by Higher Education Institutions, 2019 - 2020

  • Project Type: Project Supported by Higher Education Institutions
  • Begin Date: April 2019
  • End Date: September 2020

Project Abstract

Nowadays, robotic systems are used in autonomous (unmanned) task applications. Controller, communication module, battery, actuator, sensor, structural body, mechanical and electronic components that connect them create a robotic system. Unmanned Aerial Vehicles (UAV) can be given as an example of robotic systems. With the help of sensors attached to them, UAVs may have features such as obstacle avoidance, object recognition and target finding in military and civilian applications. Sensors can be compared to sense organs The sensors of the system enable the acquisition of the data from the environment in the execution of the scheduled tasks. Camera is a device that is used to capture environmental images. Images captured from the environment are used in tasks such as the detection, recognition and interpretation of objects with predetermined characteristics. In order to perform these type of tasks, algorithms, such as deep learning, machine learning, are widely used as image processing techniques.
Developing of decision-making processes is an important task. Artificial intelligence-based applications contribute the development of decision-making processes of autonomous systems. High processing power is needed to operate for these types of applications. In order to meet the required processing power, parameters such as performance, cost, energy consumption are taken into consideration when selecting the ideal processor. High processing power, low cost and low energy consumption are among the factors that increase efficiency.
Mini-computers are used in autonomous applications such as control of UAV, collection data from the environment, interpretation and decision-making. Products such as Raspberry Pi, Edison, BeagleBone etc. can be given as examples of mini-computer systems. Processor performance affects in direct proportion to the working speed. If the processor speed of computers where deep learning applications are run is low, the algorithm codes are executed more slowly, so the algorithm scanning period is prolonged. Equipment such as the Movidius Image Processing Unit (VPU), which can be externally connected to a computer, provides additional processing power to the used computer for decision algorithms. Because they run the codes for deep learning applications on themselves, the load on the computer main processor is reduced. In robotic applications where energy consumption is critical, such as UAVs, equipments like Movidius is preferred because of their low power consumption and increased performance in image processing. Nowadays, Python, C ++, etc. programming languages are widely used for the software side object detection, target recognition applications with the deep learning methods. Also, there are open source deep learning libraries such as TensorFlow, Yolov3, OpenCV, etc. are available for use on these language platforms.
In the proposed project work, a system will be designed to determine location of the current Global Navigation Satellite System (GNSS) of distant selected object as a target via the UAV. A gimbal system will be positioned towards the designated target with the help of the image that captured through the UAV. The distance and orientation angle data obtained from the camera and laser distance measurement module that connected to the system end will be used to calculate the GNSS coordinates of the target. When calculating the target GNSS coordinates, instantaneous measured (5 Hz, 5 times in seconds) coordinates of the UAV will be taken as basis. The performance of the developed calculation algorithm will be measured by comparing with the data obtained from the GNSS receiver which is placed on the target.