summaryrefslogtreecommitdiff
path: root/changes.txt
diff options
context:
space:
mode:
authorMartin Miller2017-03-24 13:30:20 -0500
committerMartin Miller2017-03-24 13:30:20 -0500
commita62bc5ac3429e3ae2b0d5f62e498abd8006bc98f (patch)
treeb7bfac8f3913fbb5150e2f891355c9f422ba7ff3 /changes.txt
parentdb3270767af3e5597b5a27d04083f5804743e32b (diff)
downloadrefslam-a62bc5ac3429e3ae2b0d5f62e498abd8006bc98f.zip
refslam-a62bc5ac3429e3ae2b0d5f62e498abd8006bc98f.tar.gz
Changes to changes.txt
Diffstat (limited to 'changes.txt')
-rw-r--r--changes.txt15
1 files changed, 11 insertions, 4 deletions
diff --git a/changes.txt b/changes.txt
index b1cb6c0..b21f922 100644
--- a/changes.txt
+++ b/changes.txt
@@ -12,15 +12,22 @@ and compose S as block diagonal for speed improvement.
Use quaternion reflection instead of Householder Transformation for determining
the predicted reflection measurement.
-Below here is not yet implemented:
-
-Add initial reflection view.
-
Junho sets all features to a fixed initial depth e.g 10m, 20m. I calculate the
initial depth from the reflection information. Better initialization should give
better results, because we can decrease P0 and rely on vision more. According to
Davison, setting P0 properly is important.
+Use least square to determine initial depth.
+
+Below here is not yet implemented:
+---------------------------------------------
+
+Add initial reflection view.
+
According to Davison, measurements near the center of the image (the principal
point) are more accurate than those near the edges. Thus, we can adjust R
according to the distance.
+
+Using the intitial depth estimate we get some simple sanity checks: is the
+feature above the water? Is the feature a reasonable distance away (greater than
+some minimum distance, closer than some maximum)?