Use Align function on a segment of the depth image #6113
Comments
|
In applications with a processing burden such as align, you can offload some of the work onto the GPU. With Nvidia GPUs this can be done by building the SDK with CUDA support. For non-Nvidia GPUs, vendor-neutral GLSL Processing Blocks can be used, though it may be ineffective on low-power devices. In regard to isolating part of the image, this could be done by defining a 'Bounding Box' area on the screen that excludes data outside of the box. More information on this can be found here: |
|
It is certainly possible to define multiple bounding boxes on-screen at the same time in real-time. An example of this is the MobileNet SSD person tracking program, whose multiple-box ability is demonstrated in the link below. I would therefore speculate that 32 boxes on screen should be possible as long as your computing hardware can cope with the processing involved in keeping track of the individual boxes (though the calculations may be less with static zones than with moving boxes). As long as you do the align operation after the 32 boxes have been updated then I don't see why align would not work, as the align would probably just be doing a normal align of the image data that is left remaining after the bounding boxes have excluded the unwanted parts. If you will be making use of post-processing filters then align should be done after the post-processing filters are applied in order to help avoid a type of jagged-line distortion called aliasing. |
|
This case will be closed after 7 days from the time of writing this message if there are no further comments. Thanks! |
We need to find the depth of tags identified in the RGB image. Using the Align function on the entire IR image is resource consuming, and we are trying to reduce the burden of this process. Is it possible to define a range in the IR image and run the align function of that range alone (e.g. a range of 20X20 pixels)?
The text was updated successfully, but these errors were encountered: