I-SOBOT SOCCER Padmashri Gargesa Intelligent Robotics I I (Winter 2011)
Feb 22, 2016
I-SOBOT SOCCERPadmashri GargesaIntelligent Robotics I I (Winter 2011)
Overview Objective Project Description
Environment Setup Color filtering and object detection Trajectory Planning
Links and References
Objective Install an overhead camera and calibrate
soccer field position and orientation. Determine soccer field co-ordinates through
camera vision. Determine Goal, Ball and Bot position through
overhead camera vision. Determine feasible shot region. Plan trajectory. Issue IR commands to the ISOBOT
programmatically in order to traverse the planned trajectory towards the ball
Takara Tomy’s ISOBOT 1.3 Megapixel Gigaware webcam USB UIRT – IR Transmission
Hardware
Software OpenCV USB UIRT device library
Enviornment setup Overhead webcam setup (mounted
on the ceiling) overlooking the entire field.
Soccer field 53’’ X 45’’ in dimension 66’’ vertical distance from the overhead
webcam. Black background to make other objects more
conspicuous Green color border to determine field dimension
and co-ordinates through camera vision.
Color filtering and Object detection
Objects were selected across a wide range of colors to set them apart on the color scale.
Bot detection with a green-red tiled pattern on the Bot’s head.
Object ColorField border GreenBall PeachGoal White
Field Co-ordinates Input Image is background with green border with no objects Conversion to HSV and thresholding with below values
50 < H < 180 170 < S <256 50 < V < 180
Hough lines to detect field coordinates. Rough ROI got from above set on input image and processing
is continued. Conversion to grayscale OpenCV “Contour detection” and “bounding boxes” approach
to get precise co-ordinates and field dimensions. Once field coordinates are set, border is removed. Considered using affine transformations through rotation and
warp matrices.
Object detection Ball Co-ordinates
Conversion to HSV and thresholding with below values.
6 < H < 35 35 < S <256 110 < V < 256
OpenCV “Contour detection” and “bounding boxes” used to get ball dimensions and coordinates.
Goal Co-ordinates Conversion to Grayscale OpenCV “Contour detection” and “bounding boxes”
used to get goal dimensions and coordinates
Object detection Bot Co-ordinates
To detect rear red tile Conversion to HSV and 2 levels of thresholding
Level I 0 < H < 6 84< S <256 84 < V < 256
Level II 170 < H < 200 84 < S <256 84< V < 256
The resulting 2 images are added. To detect front green tile
Conversion to HSV and thresholding with below values. 6 0< H < 100 84< S <256 84 < V < 256
OpenCV “Contour detection” and “bounding boxes” used to get bot location and orientation.
Trajectory Planning
Actual image output from the program is as shown above. Bot location and orientation is shown by the blobs on the far
left. Line connecting goal to ball is ideal strike line. Triangular region behind the ball is the feasible shot region.
Trajectory Planning
Trajectory planned is similar to a cosine curve. The curve like path is essential to make up for the inability
to control bot servos and move bot along a desired angle to a desired distance and for having to rely on the pre-programmed ISOBOT commands for BOT motion.
Goal Coordinates
Ball Coordinates
Shot Region
Bot Coordinates
Trajectory
Link and References http://www.youtube.com/watch?v=SUIOWowloTk http://opencv.willowgarage.com/wiki/ http://www.usbuirt.com/ http://www.academypublisher.com/proc/wisa09/p
apers/wisa09p267.pdf