loading
 
AVM Navigator variables
Brian from Denmark  [3 posts]
13 year
Im not able to find the email adress of the AVM Navigator creator, so i will adress my question here.
I would like to know if i can get the recognized object as a variable when i run in object recognition mode ? I would like to know which objects is in the fieldof vision. I can only find navigation related variables.
regards
Brian
EDV  [328 posts] 13 year
Brian,

Unfortunately but there is not possibility to get information about recognized objects yet. I plan to add this feature (recognized objects as array) to soon.

Dmitriy.
EDV  [328 posts] 13 year
Navigator plugin was updated. Now you can redownload the RoboRealm package for getting changes.

There were added new variables:

-== Retrieve the recognized objects ==-
*Structure consist of one int:
NV_OBJECTS_TOTAL  - total number of recognized objects
NV_ARR_OBJ_IDX    - index (identifier) of object
NV_ARR_OBJ_RECT_X - left-top corner X coordinate of recognized object
NV_ARR_OBJ_RECT_Y - left-top corner Y coordinate of recognized object
NV_ARR_OBJ_RECT_W - width of recognized object
NV_ARR_OBJ_RECT_H - height of recognized object
NV_ARR_OBJ_SIM    - similarity rate (0...1) of recognized object

*Structure consist of two int:
NV_ARR_OBJ_CORNER - left-top corner coordinate of recognized object (X,Y)

*Structure consist of eight int:
NV_ARR_OBJ_RECT   - rectangle area of recognized object (X1,Y1,X2,Y2,X3,Y3,X4,Y4).

See also attached robo file: NV_Object_Var.zip
NV_Object_Var.zip
fred miller from United States  [3 posts] 13 year
when was this updated,i bought the plug-in about a week ago
Anonymous 13 year
Fred,

note the time on the previous post. This was updated just a day ago. If you want that update just use your download link to download the latest copy. (Something you may want to keep handy given the frequent updates of RR).

STeven.
EDV  [328 posts] 13 year
Navigator plugin was updated and new variables was added:

-== Module control variable ==-
NV_IN_SET_MODE   - the current setting of module's mode: 0 - Object recognition, 1 - Navigate mode,  2 - Nova gate mode

NV_IN_WAY_NUMBER - the current route number (from 1 to 7)
NV_IN_SUBMIT_WAY - submitting of route number (value should be set 0 -> 1 for action)
NV_IN_CHECKPOINT - the current setting of checkpoint in Nova gate mode. It also needs to set 0 -> 1 for action.

See also attached robo file: NV_Control_Var.zip

NV_GATE_HORIZON - indication of gate horizon (float from -1 to 1) which shows the direction to adjust the robot's motion (0 - no adjustment).
NV_Control_Var.zip
fred miller from United States  [72 posts] 13 year
this might be a question for the navigator creator
does different light levels (lux) or different types of lights affect your program
EDV  [328 posts] 13 year
Fred,

The AVM algorithm is invariant to illumination level but if it is allocated evenly. If source of lighting also will change position and it change shadows on the object then it really will cause difficulties for recognition process. But you can make additionally training for this object after changing lighting condition and then this object will be recognized in both lighting condition (with changing of shadows too).

Dmitriy.
EDV  [328 posts] 13 year
Also algorithm AVM automatically makes additionally training on the objects within object tracking (learning is indicated by the flashing of light blue rectangle around object).
fred miller from United States  [72 posts] 13 year
it looks like it will take up resources and slow it down some,since LUX levels changes from room to room ,but mostly outside will be the biggest change,on a robot design i am working on,i am using a LUX circuit to measure light and compare with light LEVEL from a good image under ideal lighting condition,then adjust the light as needed,using X10 modules
a main reason for this is when room is dark,my design will turn on the light module and adjust light as needed,i think roborealm doesnt have a plug-in for X10 modules ,but my LEAF design does
EDV  [328 posts] 13 year
The AVM algorithm really works better with images that have good contrast condition. But if you try previously to normalize image then you will lose some source information in image (it can intensify noises) and it cause difficulties with recognition. The AVM work with grayscale image in contrast to recognition of colored markers where color saturation depends on illumination and it allow using infrared cameras with backlight. Therefore grayscale markers can be recognized even in darkness by AVM algorithm. I think it would be better if you try to use infrared camera or additional searchlight.
fred miller from United States  [72 posts] 13 year
problem with that is i would need 2 types of camera,s one for faces witch works very well and one for navigation,with my design i can adjust it to any level or no light,tried it with one of the best software for recognizing objects SENISIGHT and it works perfect,very low errors,mostly rotation of objects
only problem with SENISIGHT is the high cost
also if you software works only with grayscale how does it recognize the same object ,but with a different color,it does doesnt come up to much ,but ?,tried it with the same ball size ,but different color with SENISIGHT and picks it up with correct lighting,but in a different room with low lighting ,like my bedroom it doesnt until i turn on my circuit
havent got a chance to try your yet,still designing and testing other sensors for my AI ROBOT design
fred miller from United States  [72 posts] 13 year
also other problem using IR camera is you need a seperate interface to usb
if you are using onboard pc ,or get IR camera with wireless ,both make it high cost,if i use my robots to go outside might add lighting
doing alot of testing on different webcam also,so far like the logitech HD series or PS2 eye camera that works very well with mods (high frame rate)
i ordered the filters and lens mods to test image quality of it
EDV  [328 posts] 13 year
You can find out more about AVM here:

http://edv-detail.narod.ru/AVM_main.html
http://edv-detail.narod.ru/Library_AVM_SDK_simple_NET.html
http://forums.trossenrobotics.com/showthread.php?t=3510&page=2

And it would be nice if you will provide some video that show the problem.
fred miller from United States  [72 posts] 13 year
then i finish testing sentisight,will be trying yours
do you use color matching or just grayscale matching
program from what i see looks very good,but also needing good object recognition too,or one that has both,
in color matching is good when there is  a green ball and red ball and in can tell the difference
EDV  [328 posts] 13 year
It would be useful for comparison with regard to AVM if you also will share your results with testing SentiSight algorithm (video, diagrams etc).

>> Do you use color matching or just grayscale matching?

The AVM algorithm use only grayscale matching and it not allow to AVM see difference between green and red balls. So, good object for recognition with AVM should has appreciable texture. Also you should endeavor so that the object occupies as much as possible the square of interest area (red rectangle) during training because for AVM anything that was placed in interest area will become as object (background also).
fred miller from United States  [72 posts] 13 year
so i need a different software for object detection with colour
sentisight does have many variables in its SDK FOR COLOUR AND RGB
maybe near a hundred in the 400 page manual,but since the cost is very high over $300,someone in our group is  making a low cost and very good object recognition, she said about the same thing she uses greyscale matching too,so looking at roborealm modules i found one that does colour matching,colour statistics,plus a few other modules,then using leaf API MODULE i can have colour detection
and just use your for navigation
i did find out the lighting is a problem,sunlight shining in from a window on a object or face,or low lighting in like bedrooms or dark rooms like hallways
so with $15 circuit able to control lighting and adjust it using X10 lighting modules,also trying different webcams like microsoft hd lifecam cinema that has very good results
EDV  [328 posts] 13 year
Can you provide source video that would be used in recognition with AVM?

I would try to test it and could provide some recommendations how to process it.
EDV  [328 posts] 13 year
>> My robot seems to follow gates if they are in a straight line.
>> How much of a turn can be made before the gates are lost?

The route bends require more attention. You should walk through it several times in "Nova gate mode". You can stop the route recording if you switch mode to "Recognition mode" and back to "Nova gate mode" and then you can return robot at the start position and repeat again route/bend training. Also you can just move backward and forward along route bend when route recording is activated.

The gate weights will be adjusted when you will be training route repeatedly and significance for each route gate in AVM tree will been revaluated.
EDV  [328 posts] 13 year
Navigator plugin was updated and new variable was added:

NV_CHECKPOINT_NOW - indicates the checkpoint number when the robot arrives at the end of the current route (1-7). This variable is 0 if the checkpoint has not yet been recognized.
Autobot from Malaysia  [1 posts] 12 year
In a dynamic environment, eg People walking by as well as unexpected stationary object in the trained path? Can avm cope with this? I need to confirm this be4 I buy the avm module. Thank you in advanced
EDV  [328 posts] 12 year
* What your robot see and how it affect to navigation?

In our case the robot uses only limited sequence of images from camera for navigation. The AVM Navigator application just tries to recognize images so that be able to understand robot location. And if you show in route training process of some image sequence (just lead your robot in "Marker mode") then in automatic navigation mode the robot should have possibility to see the same sequence of images without anyone changes. Otherwise AVM Navigator will not be able to recognize images and it will cause to fail in localization and navigation.

See more details about tuning of AVM Navigator  here:
http://www.roborealm.com/forum/index.php?thread_id=4246#

Route training and navigation by map:
http://www.youtube.com/watch?v=qVz9iBazqug

This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index