search

roboticslab-uc3m / vision

Vision processing
https://robots.uc3m.es/vision/
11 stars 7 forks source link

Merge haarDetection2D+colorRegionDetection2D programs #79

Closed jgvictores closed 4 years ago

jgvictores commented 6 years ago

Merge haarDetection2D+colorRegionDetection2D programs

davidvelascogarcia commented 4 years ago

Regarding #79 issue i´ve been trying to merge colorRegionDetection /colorRegionDetection2D and haarDetection / haarDetection2D. First, i´ve started trying to merge colorRegionDetection2D with haarDetection2D. Updates in vision fork , haarColorRegionDetection2D branch. Right now works correctly at the same time and publish the results in two different ports haar and colorRegion ports.
I have to check features ports, it is added but i haven´t test yet. Picture test attached.

testhaarColorRegionDetection Figure 1. Test haar and color region detection

Regarding the name, i didn´t know the name to be selected and now i have named as haarColorRegionDetection2D, as fusion´s name. @jgvictores what do you thing about that?

PeterBowman commented 4 years ago

Is it worth in terms of efficiency having this app perform two different tasks at once?

jgvictores commented 4 years ago

Nope! @davidvelascogarcia and I have spoken regarding the proposed architecture. It would look more like the following:

photo_2019-11-11_17-09-13

PeterBowman commented 4 years ago

Looks nice. I'd keep those libs embedded within the haarColorRegionDetection2D application, unless you're really sure this could be highly reusable (given the heavy YARP constraints, I'm not sure about it). I presume that switching between modes is going to be done on init via YARP option?

jgvictores commented 4 years ago

I was not thinking in keeping the haarColorRegionDetection2D name. My original idea intended to be able to switch to tensorflow too.

Yes, I guess the config will involve passing a YARP option. The API will involve passing a YARP image and getting the features in return.

davidvelascogarcia commented 4 years ago

I have been working with proposed architecture in switchDetection2D branch. The name that i´ve selected provisionally is switchDetection2D.

switchDetection2D module, merge vision detection2D programs. The module merge colorRegionDetection2D, haarDetection2D and tensorflowDetection2D. This module works with segmentorThread. Using vision detection2D programs structure. Build an object of selected detection mode, the mode process the image received and return the processed image and results, to be published. PR at #78.

The process diagram attached in Figure 1. swicthDetection Figure 1. Diagram

Now i´m going to add Transformation class to build the selected detector objects using parameters objects instead of each parameter individually.

jgvictores commented 4 years ago

Edited description because it was way too ambitious. Enough has been done at https://github.com/roboticslab-uc3m/vision/commit/7497b4b07688aafa101889f5285abddc1c16c427 (see commit message for more tasks that were accomplished in the process):