Sensor selection

1. Webcam for feature recognition and localization



Figure 2. Logitech Webcam C210            
(picture from http://www.logitech.com/en-ca/notebook-products/webcams/devices/7022)

A conventional Logitech Webcam C210 will be mounted at the front of the UAV and will be responsible for taking images of the surrounding environment. After the images that are taken from the webcam are properly processed, the UAV will be able to recognize specific geographic features (e.g. coloured window) and able to identify its location relative to the identified feature (localization).


2. Sonar sensor for altitude control


Figure 3. Sonar Rangefinder SRF08
(Picture from http://www.acroname.com/robotics/parts/R145-SRF08.html)

A sonar sensor SRF08 will be mounted at the bottom of the UAV for the sole purpose of tracking its altitude. The specification for the sonar sensor is summarized below:

Specifications
Voltage	5 V
Current	15 mA Typ. 3 mA Standby
Frequency	40 kHz
Minimum Range	3 cm
Maximum Range	6 m
Max Analogue Gain	Variable to 1025 in 32 steps
Connection Type	I2C
Light Sensor	Front facing light sensor
Timing	Fully timed echo, freeing host computer of task
Echo	Multiple echo - keeps looking after first echo
Units	Range reported in µS, mm or inches
Weight	11.3 g (0.40 oz)
Size	43 x 20 x 17 mm (1.69 x 0.78 x 0.67 in)

Table 1. Sonar Rangefinder SRF08 Specification
(Retrieved from http://www.acroname.com/robotics/parts/R145-SRF08.html)

Mechanical frame design

A new mechanical frame had to be designed to mount the two sensors (sonar sensor and a webcam) onto the UAV. A new mechanical frame design is outlined as below:



Figure 4. Mechanical Frame Design of The UAV

software & algorithm  design

1. Feature Recognition

As for the feature recognition, it was decided that the UAV must be able to distinguish a blue coloured window frame from other surroundings, while identifying the four corner points of the window.

This feature recognition could be achieved by implementing a series of image filters as listed.

• Colour filter identifies a blue-coloured objects
• Canny edge detection detects the edges in the image
• Houghline transform comes up with four straight lines which represents the window frame based on its voting mechanism
• Harrris corner detection method was used to identify the four corner points of the window.

The result of the image process after the implementation of above filters is as shown below:


Figure 5. Image on left shows the original image taken from the webcam, while the right side image represents the filtered image after above three filters are applied. In this image, the system was successfully able to identify the blue-coloured frame with its four corners.


2. Localization

To identify the UAV's location relative to the recognized feature (blue-colour window frame), a 4 by 4 camera matrix will be applied to the image taken.

Based on the know focal length of the webcam and the known dimention of the window, and with the application of the camera matrix, it is possible to figure out pitch, yaw, roll angle of the webcam and its distance relative to the window, which can be used to figure out the relative location of the UAV in repect to the window.


Figure 6. Simple diagram showing how camera matrix works



3. Altitude Control

During its operation, we would like to keep the UAV to be flying at constant altitude in a stable manner. To do this, an extended kalman filter will be used.

An extended kalman filter accounts for nonlinearities in the motion and measurement model. Thus, this filter is suitable for uses in non-linear system such as in altitude control application for the UAV, which is expected to be non-linear and quite volatile.

The model summary for extended kalman fileter is shown below.

Motion Model:


Measurement Model:


Linearization:




Figure 7. Simulation result from the application of extended kalman filter

Hardware/software interface  design

The above elements that are discussed are organized into the overall hardware/software diagram as shown below:


Figure 8. Hardware/Sofware Interface Diagram

The bottom level corresponds to sensors that feed in new data to the vehicle for state estimation. Gucstix and MCU does the algorithm and system calculation.  OpenCV and ROS is open project code that is read through to determine the appropriate functionality.