Adapting Commercial Drones For New Industrial Security Purposes
Adapt AI for security in a commercial drone to showcase its capabilities and task performance, demonstrating how existing drones can be repurposed for new applications. Implement AI-driven user interaction by executing predefined tasks and analysing collected data to assess overall performance.
This paper outlines how pre-existing commercial drones from the market can be reused, with in the context of an indoor environment, by utilizing and adapting artificial intelligence (AI) for the purpose of completing a task that would be expected from a more advanced and expensive industrial drone. The goal of the project is to show how to reuse inexpensive commercial drones with a plethora of tasks for the drone to participate in. It will be created, and tests will be carried out to outline how the commercial drone performs in each task, using networking to interact with the drone, based off the drone’s sensors, using a Tello Ryze, a commercial drone with an HD 720 camera, 2.4 HGz 802.11n Wi-Fi and Intel 14-Core Processor. The results will be achieved by segregating performance, accuracy, and by focussing on dependent, independent and control variables.
Background to the project
The landscape of the modern technology has become ever more accessible as the technology advances. The commonality of now using AI and robotics has become a normality in everyday life for western civilizations. From this - a new marketplace for drones have arisen. Used for beautiful images, from angles that normally would be difficult to from normal photographers, to becoming a modern pastime for people to fly around and play with, has created a commercial marketplace. These kinds of drones can be acquired for around £100.
Industrially, a wide variety of uses are utilized for implementing drones into new sectors of industries. More advanced drones are being used in warfare, medicinal, agricultural and for shipment industries. These drones tend to be more expensive with higher capabilities and components, especially when implemented with AI to create a hands-off system. Industrial drones can be seen selling for over and around £10,000 per drone.
Security systems that are used for safety and well-being are mainly secured by security cameras. This amount of security in day-to-day life is ever expanding. This type of security though comes with its own constraints and limitations which will be manoeuvred around into a new modern solution for these issues.
Bringing cheap commercial drones and security cameras together we find a real niche in the security sector where drones could be adapted for safe and reliable measurements.
Project research question
Adapt AI for security in a commercial drone to showcase its capabilities and task performance, demonstrating how existing drones can be repurposed for new applications. Implement AI-driven user interaction by executing predefined tasks and analysing collected data to assess overall performance.
Project Objectives
Objective 1 - Design and create an AI feature that controls the drone’s movement:
• Understand the selected commercial drone’s capabilities.
• Design a system that recreates automated flying.
• Implement the design.
• Optimise for performance.
The first objective focuses on creating an AI feature that controls the drone’s movement giving a hands-off effect. Once understanding the commercial drones’ capabilities, the first steps of manually 7 controlling the drone can be accomplished. Implementing a design where AI can control this movement is vital for autonomous security.
Objective 2 – Design and create face recognition feature based on the drone’s camera:
• Understand the fundamentals of face recognition.
• Design a system that can accurately identify people.
• Implement the design.
• Optimise for performance.
The design of the face recognition feature is a large part of objective two. Understanding the fundamentals of face recognition and how it can perform based on different factors will help make a design of a system that can identify faces. Following that will be the implementation of the design where the performance gets optimised after to give a reliable and consistent result.
Objective 3 – Merge the two features:
• Synchronize the two features into a single system.
• Optimize performance for both.
Merging these the two features specified in objective 1 and 2, is the purpose of objective 3. Synchronising the AI drone movement with the face recognition into a singular system the optimizing of performance can be focused upon.
The finished product was designed to allow the drone to move and adjust based on a person's position, recognizing their face in an indoor environment with adequate lighting. This was accomplished through a modular and adaptable system, spanning from computer networking to the drone itself, enabling functionality across various settings.
Functional requirements
Interfaces
The interfaces in this project are designed to be accessible for both users and developers, with two main interfaces implemented.
The first interface, a Python-based system, provides real-time instructions and mechanics as they are integrated into the operation. It displays user identities included for recognition, instructions sent to the drone, and any responses received. Additionally, it presents the accuracy configuration of each identified face along with the corresponding name.
The second interface features a live-streamed view with an overlay that activates when a face enters the field of view. Initially, a “Loading…” message is displayed until the face is identified, at which point the name is updated. This interaction demonstrates the successful implementation of the architecture.
Drone Movement.
To ensure flexibility and modularity in drone movement, the code is designed to correlate positioning and motion, keeping a face centered within the field of view. The architecture relies on networking to communicate with the drone using commands. By creating a virtual representation of the drone within the computer, commands are calculated and sent out to the physical drone. Due to hardware limitations, the software does not wait for responses from the drone, allowing for faster movement.
The approach shares similarities with Software-Defined Networking (SDN), where a central entity controls the operations of an external sub-entity. By incorporating SDN principles into network communication management, the system achieves greater efficiency and adaptability in performance and monitoring. While SDN is widely used in industrial applications for drone swarms, this project is designed for a single drone rather than a swarm-based implementation.
Facial Recognition
Low Accumulated error
For the same consistency of high levels of flexibility and modularity, the code for facial recognition was designed in such a way that anyone from any race with any special facial features could participate and be identified successfully. With an ethical approach, with diversity at the forefront of direction, filtering the frame with colour filters would be unethical and inefficient. Thus, these techniques are not utilised within this project.
High Accumulated error
If these 3 diagrams were representation of frames for facial recognition in this system. We can have a good assumption that the less of the front of the face is visible, the more accumulated error in that frame there will be.
The trained data set given for this system was all based on the front of the face. Thus the gap in the system.