Bioloid Vision Module
How about a unit without a casing?
by acy.stapp
Sat Aug 25, 2007 5:41 pm
What kind of casing options are you going to have? I would most like an official casing that fits the brackets and is the same durable plastic but I don't want to waste a useful servo for it. It seems like 3d printing or custom molding a casing would be too expensive. Barring that, I'd rather have a kit and just use a plastic box on a servo bracket.
Is there room in the box for other sensors, such as a mic or the builtin IR sensors or an accelerometer?
What's the price for a bare unit and one with a box?
Do you have any visualizations of the data stream that you could share? Or a data dump that I could try visualizing?
Is there room in the box for other sensors, such as a mic or the builtin IR sensors or an accelerometer?
What's the price for a bare unit and one with a box?
Do you have any visualizations of the data stream that you could share? Or a data dump that I could try visualizing?
Re: Bioloid Vision Module
by limor
Wed Sep 19, 2007 10:21 pm
hamid_m wrote:
Hamid,
I have a few questions about the functionality of the module:
1) can you define a contiguous/non-contiguous area/shape/pattern with same/different color using PC software?
2) does each such pattern have an ID (like 1.. 70) and the returned packet from the module contains 1 or 0 if the pattern was recognized ?
2.1) or does the module return further information regarding the pattern such as probability of match, size, rotation ?
3) does the module recognize the pre-defined patterns as soon as the whole pattern is in view ?
3.1) how close (size, light condition, rotation..) should the shape be to the saved shape in order for the recognition to be positive?
4) is shape recognition size+rotation invariant (does the module do FFT on the picture ?
- Frame rate: up to 10 FPS
- Resolution: 160*120 color
- Built in color lookup table
- Image noise filtering
- Region growing for up to 255 different color categories
- Detection of up to 16 color regions in the Image
- RS485 interface, can be connected to any type of microcontroller having
- USART or directly to the Bioloid RS485 Bus.
- Output Format: 256 bytes after processing of each frame, containing information about the found color regions on the image, including coordinates, dimentions,... results can be easily used by CM5 programmed using C.
- Communication protocol: Bioloid standard communication protocol
- PC interface: 3D Color category definition, Picture Sample, Access to the camera settings.
Hamid,
I have a few questions about the functionality of the module:
1) can you define a contiguous/non-contiguous area/shape/pattern with same/different color using PC software?
2) does each such pattern have an ID (like 1.. 70) and the returned packet from the module contains 1 or 0 if the pattern was recognized ?
2.1) or does the module return further information regarding the pattern such as probability of match, size, rotation ?
3) does the module recognize the pre-defined patterns as soon as the whole pattern is in view ?
3.1) how close (size, light condition, rotation..) should the shape be to the saved shape in order for the recognition to be positive?
4) is shape recognition size+rotation invariant (does the module do FFT on the picture ?
by hamid_m
Thu Sep 20, 2007 8:24 am
1- can you define a contiguous/non-contiguous area/shape/pattern with same/different color using PC software?
The module detects the regions of up to 255 different colors. Region detection is done with an algorithm similar to flood-fill, therefore different non-contiguous regions will be detected separately.
The word "Color" is defined as a set of several RGB or YCrCb values which belong to a suface or object. Each color is defined by marking the related points in color-space and saving them in the look up table. this process will be done using a User interface over a PC, and through the CM5. This means you use a special program that rauns over CM5 during the calibration Phase, plus a Calibration user interface which runs over a PC.
2) does each such pattern have an ID (like 1.. 70) and the returned packet from the module contains 1 or 0 if the pattern was recognized ?
2.1) or does the module return further information regarding the pattern such as probability of match, size, rotation ?
3) does the module recognize the pre-defined patterns as soon as the whole pattern is in view ?
3.1) how close (size, light condition, rotation..) should the shape be to the saved shape in order for the recognition to be positive?
4) is shape recognition size+rotation invariant (does the module do FFT on the picture ?
The module does not include any pattern recognition algorithm. Each object is defined with a set of colors. The function of the Vision module is to detect and report this color spots in the image. Each detected Region has an index, but it is not guaraneed that the same index will be assigned to the same object in two different frames.
For each detected region the following information is available and can be read with the same Dynamixel communication protocol from the device:
- Color
- Number of Pixels
- Average coordinates of the detected points
- Bounding box of the detected region
Up to 16 different color regions can be detected in each frame, 8-10 FPS. There Module is also equipped with two adjustable built in noise filters in pixel and region levels.
If you have any other questions, please don't hesitate to ask me.
Hamid.
by limor
Thu Sep 20, 2007 9:17 pm
ok, so here is what I understood:
The module identifies up to 16 contiguous color areas and returns for each such area the parameters: Average coordinate, Bounding box, #pixels, etc.
The PC program allows users to define sets of discrete RGB colors that are used by the module in search of contiguous areas.
The module identifies up to 16 contiguous color areas and returns for each such area the parameters: Average coordinate, Bounding box, #pixels, etc.
The PC program allows users to define sets of discrete RGB colors that are used by the module in search of contiguous areas.
by acy.stapp
Thu Sep 20, 2007 9:19 pm
hamid_m wrote:For each detected region the following information is available and can be read with the same Dynamixel communication protocol from the device:
- Color
- Number of Pixels
- Average coordinates of the detected points
- Bounding box of the detected region
Up to 16 different color regions can be detected in each frame, 8-10 FPS.
Is it possible to get the major and semimajor axes instead of the bounding boxes? This would be a lot more useful for my application.
by hamid_m
Thu Sep 20, 2007 11:42 pm
Hi Limor, Hi acy.stapp,
Yes, it is right. Every color category is defined as a 3D geometry in the color space. number of RGBs inside a color category is not limited(of course to the number of all possible colors).
Image is procoessed online(as the pixel data arrives) in this module. Do you have any Idea to calculate the both axes without a frame buffer?
Hamid.
Yes, it is right. Every color category is defined as a 3D geometry in the color space. number of RGBs inside a color category is not limited(of course to the number of all possible colors).
Is it possible to get the major and semimajor axes instead of the bounding boxes? This would be a lot more useful for my application.
Image is procoessed online(as the pixel data arrives) in this module. Do you have any Idea to calculate the both axes without a frame buffer?
Hamid.
by hamid_m
Fri Sep 21, 2007 11:35 am
Limor,
In future, it will be possible to have versions which can be programmed partly by user. But the current plan is to decrease the price of the module as much as possible. It is of course not possible without limiting the resources.
The design team has invested a lot of work to develop the firmware. The current firmware uses almost all of the resources on the microcontroller together with lots of software tricks wich have made it possible to perform such a complex task at a relatively high frame rate over a small microcontroller.
Of course we accept every suggestion to improve the performance of the system or to adapt it to our costummers' needs.
Hamid.
In future, it will be possible to have versions which can be programmed partly by user. But the current plan is to decrease the price of the module as much as possible. It is of course not possible without limiting the resources.
The design team has invested a lot of work to develop the firmware. The current firmware uses almost all of the resources on the microcontroller together with lots of software tricks wich have made it possible to perform such a complex task at a relatively high frame rate over a small microcontroller.
Of course we accept every suggestion to improve the performance of the system or to adapt it to our costummers' needs.
Hamid.
