Controlling Quadcopters Using Hand Gestures

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Date of Award

Spring 2018

Abstract

Quadcopters are traditionally controlled using joysticks, remote controllers, mobile apps, and computers that send commands over wireless channels such as Bluetooth or Wi-Fi. Some issues with this approach include the limitation of quadcopter control by the range of electromagnetic radiation and an increase in susceptibility to interference noise. In this project, we investigated the use of computer vision to control quadcopters. Computer vision-based methods rely on the ability of the quadcopter’s camera to capture surrounding images and use pattern recognition to translate these images to meaningful and/or actionable commands. Specifically, we proposed the use of hand gestures as instructions to control quadcopters. In our approach, there was a major challenge to overcome. This was to create a robust and reliable recognition process that included a combination of static image and video stream capturing from the quadcopter’s camera, image processing, recognition of hand and body gestures, and, finally, conversion of patterns into quadcopter controls such as takeoff, landing, and so forth. Another challenge was to ensure safety assurance. While operating, the quadcopter must maintain a safe distance from obstacles such as walls or humans in indoor environments. We relied on computer vision for this task to exploit rules of vision to infer the distance from the distortion of known objects. A set of 5 gestures was studied in this work. Haar feature-based AdaBoost classifier was employed for gesture recognition. A series of experiments were conducted to measure gesture recognition accuracies considering the major scene variabilities, illumination, background, and distance. Classification accuracies obtained with various parameters showed that well-lit, clear backgrounds and gestures within 3 ft were recognized correctly with over 90% accuracy. Limitations of the current framework and feasible solutions for better gesture recognition are discussed. In conclusion, with this project we present a novel and successful approach towards controlling the quadcopters with body language that alleviates limitations of current control methods.

Advisor

Mutlu Mete

Subject Categories

Computer Sciences | Physical Sciences and Mathematics

Link to Full Text

Share

COinS