16-08-2012, 03:00 PM
Vision-Based Autopilot Implementation Using a Quadrotor Helicopter
1Vision-Based Autopilot.pdf (Size: 831.21 KB / Downloads: 87)
Abstract
A monocular vision-based pose estimation and stabilization system for a quadrotor
helicopter is proposed. The goal of this project is to enable the helicopter to hover in place
using a vision-based autopilot. The method consists of a single camera onboard the
helicopter that is used to estimate the attitude and the position of the vehicle, except for the
altitude, which is controlled manually from a joystick. These parameters are calculated
using three dark colored targets mounted on a white wall. An algorithm processes the video
frames to extract the necessary information, which is evaluated for errors and then passed to
the control algorithm. The control signal for the helicopter is output via the audio port and
driven through a custom-made circuit to eliminate noise and convert it to the necessary
format. The results from flight tests are presented with the system’s advantages, limitations,
and drawbacks discussed.
Introduction
N recent years, an abundance of different kinds of remotely controlled (RC) vehicles created a large potential for
finding an appropriate system to convert into an Unmanned Aerial Vehicle (UAV) by implementing either
onboard or a ground based autopilot. Both methods have their advantages and drawbacks. Onboard autopilots add
weight to the UAV, which often is a serious issue for mini UAVs, but have an advantage of controlling the vehicle
even when the communication link between the autopilot and a ground station is lost. Ground based autopilots
usually control the vehicle using an existing transmitter, therefore excluding the installation of additional devices on
the UAV. The drawback of this system is that in the case of lost communication between the ground station and the
UAV, the vehicle must be controlled manually, otherwise serious damage can be expected.
Error Correction
The rectangles that are used to obtain attitude information
can be posted in many ways, as long as they are reasonable
distances apart and the appropriate distances among them are
input into the program. When the program is initialized, it
searches for these objects and assigns values starting from the
left-hand side. Figure 3 shows a sample video frame with the
rectangles numbered manually. Occasionally some of the
video frames are noisy (Fig. 4), and the software is unable to
extract the necessary information from it. In some cases, some
of the rectangles cannot be detected for several frames (Fig. 5).
To reduce the effects of these disturbances, the program
continuously monitors the positions of the three rectangles.
When any of the rectangles cannot be detected, the last valid
position for that rectangle is used until the rectangle is detected
again. If for twelve consecutive frames the rectangle cannot be
detected, the helicopter goes to its landing mode.
Controller Design
There are four controllers for the vehicle, including the yaw, roll, pitch, and position controllers. Each of these
controllers was designed and tested for performance separately, while everything else was controlled manually.
After all the controllers performed satisfactorily in individual tests, they were combined and adjusted to yield
satisfactory results. Each of the four controllers will be discussed in detail as follows:
Yaw Controller
The yaw controller is the least critical of all the controllers because it does not have any effect on the helicopter’s
motion. This was the first controller to be independently developed and tested. Furthermore, after incorporating it
with all other controllers, no addditional adjusting was required to achieve desirable control in the yaw direction. A
proportional plus derivative controller was implemented for this purpose and is shown in Fig. 9. The proportional
term, Kp, and the derivative term, KD, were chosen empirically. It is worth mentioning that during low-speed control,
the closed loop control system works well without the KD term. However, running the helicopter at higher speeds
increases the settling time, hence requiring an addition of the derivative term.
Conclusion
We have demonstrated a monocular vision-based feedback control system for a quadrotor helicopter. The results
show that this method yields acceptable results. The main advantage of this system is the fact that there is no need of
installing any additional devices on the vehicle. Moreover, the system does not use serial nor parallel ports to control
the vehicle, but instead it uses an audio port, which is widely available on all computers. However, the altitude of
the vehicle is controlled manually, which can be a challenging thing to do, because the helicopter changes the
altitude frequently, depending on its attitude. Also, the helicopter moves more in X and Y direction compared to the
systems that incorporate IMUs and gyros for stabilization. In the future, the three targets on the wall can be replaced
by a specifically designed pattern that will provide more accurate information about the helicopter’s pose.