UAV Object Detection using AI/ML

Overview

This project aims to enhance the safety and efficiency of unmanned aerial vehicles (UAVs) by implementing an obstacle detection system. Leveraging advanced sensors and computer vision algorithms, the UAV will autonomously detect its environment and where it can land, ensuring smooth and secure navigation even in complex environments. By successfully implementing this project, we aim to significantly improve the safety and reliability of UAV operations in various applications such as aerial surveillance, search and rescue, and infrastructure inspection.

Objectives

1. Sensor Integration: Integrating various sensors such as cameras to provide comprehensive coverage and accurate obstacle detection capabilities.
2. Real-time Processing: Developing real-time processing algorithms to analyze sensor data and identify objects with high accuracy and minimal latency.
3. Machine Learning: Employing machine learning techniques to enhance obstacle recognition capabilities, enabling the UAV to adapt to different environments and recognize novel obstacles.
4. Autonomous Navigation: Implementing autonomous navigation algorithms to dynamically adjust the UAV’s flight path in response to detected obstacles, ensuring safe and efficient navigation.
5. Testing and Validation: Conducting extensive testing and validation in simulated and real-world environments to assess the performance and reliability of the obstacle detection system.

Approach

The approach our team took was using a camera with integrated object detection capabilities and tuning those to fit our needs. We used this camera and implemented a custom trained model to detect the landing pad that we built. Then, using the cameras information on object detection, we implemented autonomous navigation instructions for the drone to use to navigate itself over to the landing pad and then land. Using the Jetson Nano as our onboard computer to do the computations and to hold the model information and navigation instructions. We also used 3D modeling with SolidWorks to create a camera mount on the front of the drone.

Outcomes

This project was completed well within our budget, and we saved the sponsor $400.04. This was mostly due to the inheritance of the drone from the previous team to work on this project. However, our unique approach using a camera with built-in object detection led to us getting much further in this project than any previous team. The overall outcome proved our method was effective. With more time the accuracy of the landing directions could have been perfected and we hope another team will take up this project where we left off and make the adjustments that we did not have time to do.