It seems that the bins of your histogram are 17m in width if my eyes don't deceive me. That's not uncommon for aerial surveys when detectability isn't perfect under the plane, near the trackline.
One option could be to do nothing; the fitted detection function will smooth through the "few" detections in the first bin and the "considerable" detections in the second and subsequent bins. Goodness of fit statistics will suffer a bit as a result, but perhaps
The other option is to left-truncate data, mentioned in Section 10.3.2.1 of Buckland et al. (2015) and Section 8.4.3 in Buckland et al. (2001). That could involve removing the first bin such that detectability is assumed perfect at 20m. As an aside, I would
probably also impose right truncation on the data in the histogram as well, perhaps back as far as 500-600m.
There is an example of left truncation associated with a camera trap analysis found at
this shows the syntax to be used with the
ds() function to produce left (and right) truncation.