why the error(pix) from GCP is higher than reproduction error without GCP?

[ken lau]

Number of images: 295              Camera stations: 286
Flying altitude: 7.45 m                       Tie points: 276,036
Ground resolution: 2.46 mm/pix       Projections: 924,151
Coverage area: 3.34e+03 m²       Reprojection error: 0.531 pix


Reprojection error: 0.531 pix

GCP error in pix: 0.623


after applying GCP, the error in pixel should be smaller than Reprojection error, but the result is different now.

[Paulo]

Hello Ken lau,

typically the RMS reprojection error is less than GCP err (pix) as reprojection error is the RMS error of all tie points reprojected on images where they are detected. The average tie point automatic detection error (computer vision) is much higher than manual GCP measurement on image (0.1 pix vs 0.5 pix). So your results are quite normal....

Hope this makes sense,

[ken lau]

Number of images: 309   Camera stations: 309
Flying altitude: 10 m   Tie points: 96,143
Ground resolution: 3.28 mm/pix   Projections: 666,699
Coverage area: 2.58e+03 m²   Reprojection error: 1.09 pix


other case showed different result.Reprojection error: 1.09 pix, while the image pix is 0.634. in fact, what should GCP's image pix be? 0.5 pix is the perfect?



thank you for answering my question

[Paulo]

Hi ken,

i would suggest that 1.09 pix RMS repro error is a little high. Then you can do gradual selection on tie pts to eliminate points with high repro error or Projection accuracy to eliminate points that are probably erroneous. 0.5 pix for marker accuracy is a good assumption as it is difficult to localize (manually) a GCP on image at better than 0.5 pix accuracy.