AR.Drone Helicopter Optical Flow Test #1
Uploader Comments (ytsutano)
Top Comments
-
Ok I understand that you're in college, I also understand you have a helicopter with a video camera on it, I even understand that you are using (persumably) OpenCV's motion tracking algorthims.
What I don't understand is why are you are flying it in the hallway when the obvious place for this technology is the women's shower room.
All Comments (15)
-
and skynet was born....
-
It looks like you're looking for new features on every frame, which usually ins't necessary, A simple thing to try is to just call goodFeaturesToTrack on every Nth frame, and call calcOpticalFlowPyrLK every frame (recycle the points in between Nth frames). This will reduce CPU utilization and you should see more interesting flow detection. More complex methods maintain a pool of active points. Points 'die' based on some criteria then get replaced with new ones as they get killed.
-
Did you use the Lucas-Kanade Pyramidal implementation in OpenCV or Horn Schunck? What do you use the left-avg and right-avg for? is it just meant for obstacle avoidance or also to derive the egomotion? It looks awesome!
-
@TheOnlyDeathAdder Learn C++ and some OpenCV Thats the most Used programming language.
-
give it a pellet gun and the capability to do a barrel roll thne after it suceeds in those fields give it a RPG and attach a search light and a laser dot have the laser set to follow moving objects
-
I would love to start programming in real world objects, where should i start?
-
You should've used your ray-gun on the guy that's running.
You should add compensation for self motion like the human stereo vision does. When our head turns left it compensates for that so we don't think that everything else moves right. The vellocity vectors we see should be affected by the self's motion vector. Maybe you don't care about that but saying not to aggrevate people.
stylianyt 3 months ago
@stylianyt Good point. My approach here is to minimize such processing at this sensing level. The information obtained from vision, accelerometer, gyros, etc. are combined at a higher level component (brain), and this is where the compensation happens. This is to keep generality and flexibility of each component while maximizing the overall system performance.
ytsutano 3 months ago