Abstract: This paper proposed a Convolutional Neural Network method for a Quadcopter to detect and track human motion. The proposed methodology is constructed based on computer vision combined with a Convolutional Neural Networks (CNN) model and a Proportional Integral Differential (PID) controller. The Convolutional Neural Network (CNN) model is used for detecting humans, with input data being human images, through the structure of Convolutional Neural Networks and output data being a predicted bounding box of the people. In addition, the Centroid tracking method is responsible for calculating the offset between the center of the bounding box created from the Convolutional Neural Networks and the center of the frame to control the flight direction and ensure that the Quadcopter correctly tracks a human. Furthermore, the PID controller has the purpose of calculating the control values of the Quadcopter to maintain position, ensure speed and stability when flying. The superiority of the SSD-MobileNet model is validated through fundamental evaluation criteria in various identification scenarios. The effectiveness of the proposed control approach is supported by experimental flight and signal charts. Quadcopters are specifically designed to withstand the influence of the surrounding environment and fly outdoors. The experimental results demonstrate that the flight model meets the control methods and quality assurance requirements throughout the entire flight.

Human Detection, SSD-MobileNet, CNN, Centroid Tracking, Quadcopter, PID

Members

POSTER