When there are multiple (depth-ordered) layers to keep track of. Detector, track pointer, perceiver, monitor, and reporter are coded to operate on signals generated. More...
When there are multiple (depth-ordered) layers to keep track of. Detector, track pointer, perceiver, monitor, and reporter are coded to operate on signals generated.
Layered detector, track pointer, and perceiver classes for surveillance of a process with various objects or instances in it. The code decomposes the sensed image into different scene elements. For simplicity of testing and development, some of the code here may involve layers but not necessarily recover all of them. Rather, it will be focused on extracting a single layer and ignoring the other ones.
Given how python works regarding code import, these implementations are typically self contained within a single file (python package per doyxgen, if you will). Though longer, there is a single python import to manage and all co-compatible elements of the monitor are in the same code file. If done properly, the individual classes will lean heavily on other code libraries and be relatively compact with only the major differences coded up.
What should be contained in this file would be:
- Layered detector from RGBD input.
- Layered trackpointers based on layered detector output.
- Layered perceiver that combine detector + trackpointers.
- A calibration scheme for the entire process with saving to YAML and HDF5 files.
Note: There is older code that broke everything up into individual elements with ROS message passing, or some form of sequential operation. It did not work so well and has been superceded by the newer code here. The older, now obsolete code includes: 'human_seg', 'robot_seg', 'tabletop_seg', 'puzzle_seg', 'base_bg', 'base_fg', 'base'. It will eventually be removed once the new system has been shown to work well.
