Seed mapping and recognition of crop and weed seedlings with computer vision

1
Seed mapping and recognition of crop and weed
seedlings with computer vision

• Henning T. Søgaard
• Danish Institute of Agricultural Sciences
• Workshop in Spatial and dynamic weed
  measurements and innovative weeding technologies
• Bygholm, Denmark 14th - 15th of November 2005

2
Background
Danish Robotic Weeding Project

• Geo-referenced seeding with RTK-GPS (seed
  mapping, 2-5 cm accuracy)
• Recognition of crop and weed seedlings by
  computer vision
• Mechanical/physical weeding tools
• Micro-spray system for close-to-crop weed
  control
• All operations can be performed by RTK-GPS
  controlled autonomous vehicles

3
Seed Mapping
Robotic Weeding
Problem no weeding close to the crop
4
Computer Vision
Robotic Weeding
Problem high weed density ? low reliability
5
Question
Why not combine seed map information with
computer vision?
6
Seed Mapping Computer Vision Robotic Weeding
Coarse localisation of crop map data
RTK-GPS Accurate localisation recognition of
crop by computer vision
7
Recognition of crop seedlings
Sugar beet image
8
Sugar Beet Field
Micro-Spray System
Camera
RTK-GPS, compass
9
Sugar Beet Field
Centre of Field of View
a
10
Calculating the Predicted Sugar Beet Position
Stored Seed Position
Centre of Field of View
a
11
Sources for the Prediction Error

• First Pass - Seeding
• RTK-GPS St.Dev. ? 10 mm
• Tilt sensor St.Dev. ? 10 mm
• Seed displacement in furrow St.Dev. ? 9 mm
• Seed-plant deviation St.Dev. ? 12 mm
• Second Pass Weeding
• RTK-GPS St.Dev. ? 10 mm
• Tilt sensor St.Dev. ? 10 mm
• Total (?? Std.Dev.i2) St.Dev. ? 25 mm

12
Experiment Seeding
RTK-GPS antenna, tilt sensors and compass
Registration of seeds dropped
13
Experiment Image Capture
14
Experiment - Data

• Seed map
• 40 geo-referenced images
• Predicted positions of the sugar beets within the
  images
• Calculated differences between predicted and
  actual sugar beet positions

15
Results
16
Conclusion

• Combination of seed maps and computer vision ?
  efficient localisation of crop plants

17
Thank you for your attention