Use wider border in rejecting april tag detections
In reviewing logs, there were examples where there were bad corners at 5-8 pixels
from the edge, so filtering a little more aggressively so we get rid of these
Since they're also detections in portions of the image with the most distortion,
I think it's OK to be more aggressive in rejecting these
Change-Id: I65c4be00c5e1407982be2f31c97b8cf6e4aebe91
Signed-off-by: Jim Ostrowski <yimmy13@gmail.com>
diff --git a/y2023/vision/aprilrobotics.cc b/y2023/vision/aprilrobotics.cc
index 70a3f30..217d5b0 100644
--- a/y2023/vision/aprilrobotics.cc
+++ b/y2023/vision/aprilrobotics.cc
@@ -8,7 +8,7 @@
DEFINE_double(min_decision_margin, 75.0,
"Minimum decision margin (confidence) for an apriltag detection");
-DEFINE_int32(pixel_border, 3,
+DEFINE_int32(pixel_border, 10,
"Size of image border within which to reject detected corners");
DEFINE_double(
max_expected_distortion, 0.0005,