IMHO down sampling the image could greatly improve speed, though I'm not sure about the impact on detection/tracking performance.Worth a try though.
Meanwhile, I don't think the CPU of IPhone have hardware support for floating point operations. If you are using such operations to calculate SURF discriptors, I think changing them to fixed point operations might help as well, although upflows and downflows might be a little tricky to handle.
@darxide12345 Thanks for the comment! I'll check out your suggestions. Also have a look at: "OpenCV FAST corners and OpenGL on iPhone 3Gs" where I captured incoming video frames at a lower resolution and used OpenGL output for very fast FAST detection
You will need to google around a lot, but you can run OpenCV on the iPhone, you will use AVFoundation to access the camera video stream and it is not difficult, it will just takes some time.
@yxccumt Thank you! Yes, I used OpenSURF. However, I found that SURF is just too slow. I decided to use sparse optical flow for tracking (lucas kanade) instead of descriptors. What are you busy with?
@carelvwyk I saw Learning OpenCV on the corner. So you you use lucas kanade for tracking instead of SURF feature matchin? I'm working on some mobile phone head tracking stuff. I'm a newbie on phone programming ... Is running OpenCV difficult on iphone? How to access the camera video stream? In opencv it's just a function. Is it difficult on the iphone?
Now there's a Lucas Kanade Optical Flow project as well to test.
checkout labs . success-ware . com
Cheers,
Oded
odedbendov 11 months ago
Nice job,
checkout this implementation of fast corners - search "fast corner iphone" and see the hatzlaha one (success in Hebrew)
it was hand tailored to offer the best performance I could reach, so it might offer some performance gain if you could incorporate it.
Cheers!
TheFreddy4 1 year ago
Comment removed
darxide12345 1 year ago
Good job!
IMHO down sampling the image could greatly improve speed, though I'm not sure about the impact on detection/tracking performance.Worth a try though.
Meanwhile, I don't think the CPU of IPhone have hardware support for floating point operations. If you are using such operations to calculate SURF discriptors, I think changing them to fixed point operations might help as well, although upflows and downflows might be a little tricky to handle.
darxide12345 1 year ago
@darxide12345 Thanks for the comment! I'll check out your suggestions. Also have a look at: "OpenCV FAST corners and OpenGL on iPhone 3Gs" where I captured incoming video frames at a lower resolution and used OpenGL output for very fast FAST detection
carelvwyk 1 year ago
I don't think you need SURF for that.
You will need to google around a lot, but you can run OpenCV on the iPhone, you will use AVFoundation to access the camera video stream and it is not difficult, it will just takes some time.
carelvwyk 1 year ago
@carelvwyk Yes, for my job, SURF is not needed. But I'm generally interested in all vision techs. Thanks for your help!
yxccumt 1 year ago
Great job! What SURF code do you use? OpenSURF?
yxccumt 1 year ago
@yxccumt Thank you! Yes, I used OpenSURF. However, I found that SURF is just too slow. I decided to use sparse optical flow for tracking (lucas kanade) instead of descriptors. What are you busy with?
carelvwyk 1 year ago
@carelvwyk I saw Learning OpenCV on the corner. So you you use lucas kanade for tracking instead of SURF feature matchin? I'm working on some mobile phone head tracking stuff. I'm a newbie on phone programming ... Is running OpenCV difficult on iphone? How to access the camera video stream? In opencv it's just a function. Is it difficult on the iphone?
yxccumt 1 year ago
Well youtube is being a dick and not sending through my reply.
carelvwyk 1 year ago