How to identify what's new in image?
I have a camera facing into a trashbin. I want to take a picture of every new threwn item.
To start with I don't realy know what way to go. Take one picture after something is threwn and compare with last taken picture or use live camera stream analyse.
The problem is to mask that new item as some stuff like glassbottles are both transparent and heavy which makes them hard to mask and they often moves other trash little bit. If I have one white item in the trashbin and the new item is also white it's hard to mask that new item by absdiff as the differences is to small. In those cases maybe a edge-detection solution could fix it, but I don't know how to remove edges of all old stuff and only get the edge of the new item, and then mask the area inside.
Any suggestions how to solve this?
When item will land, it will probably move other items in trash bin, so just taking a difference between image before and image after might fail sometimes. I'd rather analyze picture that was taken while the new item is in process of falling.
I think you are right. The problem is to get a sharp image while the item is falling with when the camera only make 60fps. But I was thinking if it would be possible to track the falling item to where it lands and then analyse that area.
Else I could use the depth information from the camera (Using realsense SR300) but there is lot of noise from edges that has to be handled when using absdiff.
Analyzing the object when falling is hard to succeed as first of all the hand of the person who are throwing the trash will be included and then it's hard to get a sharp image. Maybe with a good highspeed camera it would be possible but don't have one.
Maybe it would be possible to follow the trash down and analyse when it's in the bin, but I guess the hand can be the problem if hides the fall.
The best would be if I could mask the new object with a before and after image. It should be possible to filter it in some kind of way as the human eye can see what's new. How do I ignore movements of already throwed trash? I have both depthinfo and rgb image before and after. The depthcamera have no problem with thick objects, but on thinner ...(more)
Any suggestions?