Student Research

Autonomous Target Acquisition and Recognition using UAVs

The responsibilities assigned to Uninhabited Aerial Vehicles (UAVs) have become increasingly complex. One such mission involves surveillance for the military. For this to happen, the UAVs must have the capability to autonomously acquire targets and recognize images. The scope of this research is to develop such a system for a fixed wing UAV. Computer based system is used in conjunction with an automatic flight control system to locate targets of interest and apply image recognition algorithms to determine the characteristics of the target. A camera mounted to a gimbal system, which is attached to the underside of the airframe and controlled by a microcontroller, servos, and an attached inertial measurement unit (IMU), provides image capture for the computer vision software. Upon capture, the image is geotagged with coordinates provided from the onboard Global Positioning System (GPS) module. The vision software then uses a grayscale filter with canny edge detection and convex hull algorithms developed from the OpenCV library, an open source library of computer vision algorithms, to separate objects of interest from the background. The software then takes unique sets of features of the images and compares them to a database for shape, color, orientation, and alphanumeric correlations. To perform autonomous navigation, an open-source autopilot called the ArduPilot Mega 1.4 is being used. Controller gains for autonomous flight are tuned in hardware-in-the-loop (HIL) simulations. Ground testing of the image recognition software is performed by moving targets below the airplane and observing the image-receipt by the Ground Control Station for the confirmation of target acquisition. The entire system has been verified in flight tests, involving simulated targets. The overall system is capable of autonomous target acquisition and recognition, while the airplane is flying autonomously.


In Collection: