Speed measurement via optical flow / visual odometry
Budget: €30 – €250 EUR
We're developing a machine to work in agricultural fields. In order work very precisely, we need to know exactly how the machine moves in time. Therefore, we calculate the distance traveled / traveling speed out of subsequent images by means of visual odometry.
The algorithm already in use has to be validated against ground truth data. Ground truth data exists in form of image sets with a physical tape measure visible in the images (see example). The 2 image sets contain of 600 images each.
Once validated, the existing algorithm has to be compared with other existing optical flow algorithms. Finally, the best performing algorithm is chosen and eventually further optimised for even more exact and reliable speed measurements.
Workflow/Tasks:
- Transform the ground truth data into computer readable form (read the tape measure in the images and add the values into a existing .csv-file)
-Validate the existing speed measurement algorithm against the human measured distances
- Implement and compare different image representations as a base for the optical flow algorithms
(right now, the optical flow algorithm gets a vegetation index image (ExGreen) to run on instead of the raw image)
- Implement different optical flow algorithms and validate them against ground truth data (to compare them with the existing one)
- Eventually further optimize the best performing algorithm
The algorithm already in use has to be validated against ground truth data. Ground truth data exists in form of image sets with a physical tape measure visible in the images (see example). The 2 image sets contain of 600 images each.
Once validated, the existing algorithm has to be compared with other existing optical flow algorithms. Finally, the best performing algorithm is chosen and eventually further optimised for even more exact and reliable speed measurements.
Workflow/Tasks:
- Transform the ground truth data into computer readable form (read the tape measure in the images and add the values into a existing .csv-file)
-Validate the existing speed measurement algorithm against the human measured distances
- Implement and compare different image representations as a base for the optical flow algorithms
(right now, the optical flow algorithm gets a vegetation index image (ExGreen) to run on instead of the raw image)
- Implement different optical flow algorithms and validate them against ground truth data (to compare them with the existing one)
- Eventually further optimize the best performing algorithm