Special Sensor: CMU Cam 3
Student Name John Kurien
Robot Name Hard 1
Instructors Dr. A. Antonio Arroyo
Dr. Eric M. Schwartz
TA Thomas Vermeer
Mike Pridgen
EEL5666: Intelligent Machine Design Laboratory
Table of Contents
Abstract3
Introduction4
Integrated System5
Mobile Platform6
Actuation7
Sensing and Behavior8
Experimental Layout and Results9
Conclusion11
Documentation12
Abstract
The designed robot will perform the function of playing paper football with a human opponent. In order to identify the football to hit it, the robot requires a visual sensing mechanism to determine the location of the football. The CMU Cam 3 will be used to perform this task because it can easily identify the football and track it as the robot moves closer to it. Additionally the hardware is well documented and has been proven to be successful by previous applications.
Introduction
The integration of the camera into the hardware of the robotic system will be explained first. Mechanical considerations will be covered next such as the mounting location of the camera and its actuation to track the paper football. The sensing capabilities and software behavior of the camera can then be determined. This will finally lead to determining a method to test the sensor and if it has a feasible implementation.
Integrated System
The CMU Cam 3 is connected directly to the Atmel Atemga 128 communicating through UART. The power to the camera will be supplied directly from the 8 V battery which falls between the 5.5 -10 V input range. Since the camera does not draw large amounts of current sporadically, there is no plan for any power protection circuitry. If the system is determined to be unstable another power solution will have to be developed.
Mobile Platform
The first iteration of the mobile platform did not have enough area to support all of the circuit hardware in addition to having enough clearance to the floor to place a kicking mechanism. As a result, a second platform design was created. The above design does not include a placement area for the CMU Camera however plans are to place it atop the robot in the front. There will be a bit of an overhang of the camera from the front most part of the robot to allow the camera to pivot (as explained in the actuation section) to look directly down on the football once the robot has approached it.
The camera will look directly down on the football to determine the orientation of the football and how off center the football is from the center of the robot. These perceived inputs in addition to the distance to the goal line will determine how fast the DC motor must spin to hit the football a desired distance.
Actuation
The CMU Camera has been attached to a servo to allow the camera to tilt (not pan) as it approaches the ball. This will allow the camera to ultimately look down on the football when it reaches it. The servo is anticipated to be controlled from the microcontroller (Atmega 128) side rather than off loading this task to the CMU Camera to do. If there is an issue of running out of PWM ports (which could possibly occur) then this will be the first PWM to be offloaded from the Atmega to another piece of hardware (the camera).
Sensing and Behavior
Since there is a controlled environment for the robot to play in, the color of the football can be hardcoded into the robot. This will allow to the robot to identify only the football and not require recalibration every time the robot is used. This intended non-calibration technique will have to be verified to work. The camera will allow the robot to sense the paper football to approach it. Calculations will have to be performed to allow the robot to come straight in line (perpendicular to the goal line) with the football to hit it straight.
Once the previous task is complete the color tracking will be implemented to track the ball as the robot gets closer to it. This color tracking is built into the CMU Cam 2 firmware which is emulated on the CMU Cam 3. This has been verified to work and is as simple as sending a 4 byte command and receiving a standard packet of data via UART saying where the ball is located in the camera’s screen.
Once the color tracking is complete to approach the ball, an algorithm will be determined to analyze the orientation of the football. This will probably be implemented with an edge detection algorithm which has commands already built into the CMU Cam 2 emulated software. The orientation of the football is of concern since depending on how the football is turns, the wheel of the DC Motor will grab either more or less of the football for a shorter or longer time of a revolution of the wheel. This dramatically effects how much further the football travels.
Experimental Layout and Results
A java based software program is included with the CMU Cam 2 to be able to visually see what the camera sees. The above frame was grabbed with the software using the emulated CMU Cam 2 software.
It is intended to be able to track the football once it is moving (when the robot is actually moving). To do this the color of the football must first be determined as shown in the above figure (left). As a test we would like to see how well the camera can distinguish colors. As a test, a piece of masking tape was placed next to the football. The tape was light brown in color which could possibly be confused with white, however it is seen that the camera was still able to identify the football. When in its environment, the robot will not see any white on the field except for the white of the football. All other colors will be green (field), orange (walls), and possibly gray (stands).
The final figure displays the actual tracking of the football as the paper is moved around. When viewed as a video, the red dot can be seen to move along with the football in the same direction. Rather than capturing this information with a java based program, the robot will received this information in a packet format stating where the white football is in the camera’s view and how reliable a tracking the camera has on the football.
Conclusion
Overall the CMU Camera 3 seems to be a very confident resolution to the issue of tracking the paper football. The actual software algorithms to effectively use the camera must still be developed, however from an abstract stand point there seems to be no reason why the camera would not be satisfactory. Further experimentation must be done on the camera to verify its functionality and reliability.
Documentation
1 | Page