Projects

Quadcopter Drone Autonomously Recognizing and Flying

In this project, we developed an advanced autonomous UAV system that integrates computer vision, embedded control.

Our system design began with the creation of a sophisticated quadcopter platform, which featured a robust vision system for real-time digit recognition and classification.

We optimized OpenCV-based image processing algorithms to process the MNIST dataset, integrating efficient memory management techniques to ensure smooth real-time operation.

To enhance classification accuracy, a convolutional neural network (CNN) was trained and deployed, achieving an accuracy rate exceeding 95%.

The integration of these features was made possible through STM32-based embedded development, which allowed us to embed the CNN model and vision system into the UAV’s control architecture.

The system was designed to use the recognized digits and Chinese characters to determine the next flight direction. Based on the recognition results, the UAV could autonomously adjust its flight path, generating trajectory commands for navigation.

This real-time decision-making capability allowed the UAV to dynamically adapt its movement based on the recognized objects, significantly improving its agility and performance in both competitive and real-world scenarios.