Here it is!
It seems the original versions of the pixycam didn’t have the ability to detect anything other than color. Newer versions have the ability to detect and differentiate simple patterns (EG: red square next to blue square) meaning that the software used in the v5 camera will be slightly different; but the basis of computation is the same. Additionally, the first versions of the pixy cam only differentiated between 6 colors: red, green, blue, yellow, cyan, and magenta. Adding specific colors was added later. Now let’s get into how the pixycam works. First, the image is split into individual color channels and subtracting the grey-scale of the image. This leaves just the brightest colors of the initial color channel. The image is then run through a filter to remove noise. Next, each pixel in the image (currently a greyscale) is compared to a threshold value; if the pixel is greater, it becomes white, if lower, black. At this point, no more significant image processing occurs, and all the detection has happened for that color channel. Anything that is white is “detected” anything that is black is just the background (hopefully). This process occurs for every color channel, yielding 6 binary black and white images which are the final result. Technical discussion go!
but the question is, can it allow me to aim lasers into people’s eyes?
Your explanation is completly incorrect for both the current pixycam and the V5 Vision sensor.
What’s incorrect about the explanation? I’m just wondering so I know what to expect from v5
Yeah what he said!!!
The explanation is incorrect, the existing pixycam does not operate that way, there’s a lot more to it.
Can you explain how it does operate?
That is not the pixycam thesis paper. Just looks like someone using a pixycam on some sort of RC car school project. “to correlate closer to the processed image of the Pixycam” (Page 11). They seem to be trying to replicate the real pixycam algorithm. Of course, they’re missing a lot so the product still has quite a bit of room for improvement.
Perhaps at some point we will, but for now you can read all about pixy (also known as cmucam5) on this site. Please remember that the V5 vision sensor is an evolution of pixy and some details of the implementation may be different.
http://www.cmucam.org/projects/cmucam5
So on that website, it says that pixy can track hundreds of things at a time, but I believe it was said that the v5 vision sensor can only track seven things at once. Is there a reason for this?
wait where did it say that?
The V5 vision sensor can learn 7 different object signatures, there may be many objects in an image with the same signature, therefore it can track many more than 7 objects.
The vision sensor can also understand object signatures that are placed close to each other and in a specified order. One of the demonstrations at CES shows an application of this, the four cards in this image are used to instruct a motor on the V5 brain to run forwards, backwards etc.
Oh ok. Also, the website says that you can use color codes, will the vision sensor be able to do that as well?
The color patterns explained above are what the pixy refers to as “color codes”.
As I said multiple times in my post and in the title, this was an early model. And I couldn’t have gotten too wrong seeing as I was just putting the code in the article into layman’s terms. Attached is all the code provided in the article
@antichamber again, this is not real pixycam code. And it’s not an early model. This is a student’s version of an attempt to emulate the pixycam algorithm. The real algorithm shares some similarities with what you’re describing, but it really is an extreme oversimplification.
Let’s say for instance, I was “completely incorrect” If that were to be the case, not only would the creators of the pixycam have thrown away all of their development, but you would be counteracting official statements from vex saying that it’s a “hue based object tracking system” which if you have read my explanation, or the thesis I linked, you would have noticed that the system’s detection system is completely based on hue. Am I wrong in believing that a system that tracks objects based on their color is in any way similar to a system that tracks objects based on their color and or simple patterns of color?
Being an early kickstarter supporter of the PixyCam, and prior CMUCams… @jpearman is right this is entirely incorrect, given even the initial information I had from the Pixycam.
I was able to do object detection with the early Pixycams and store color signatures. And I’ve been able to add color codes from the start.
I have a PixyCam (you can get some amazing things via Kickstarter. ) it does colors pretty well. It’s wicked fast for some things. I think that the 2019-20(*) game will have elements that will benefit from having the sensor. In my usage (and I think in all cases with digital sensors) the lighting makes a difference.
Only if the lasers are attached to sharks. Unlikely to have water in the 2018-19 game or the 2019-2020 game, so you may need to wait on this.
(*) While VEX would love instant adoption of all teams going to V5, they are pragmatic. Because of costs of going to V5 and limited team resources, they understand how the rollout goes, since they have done this before. I don’t know of any other company that goes “Hey, you just bought this, we are going to buy it back to help you get the new product” the way that VEX does. Buy an iAnything from Apple and try that. I think Paul’s influence on RECF will have them not release a game until most of us can afford, get, test and use the new vision sensor.
As someone that used the original CMU cam on the 0.5 PIC system I couldn’t be more thrilled about this upgrade. Lots of you are going “pfft, I want full motion dis-assembly and recognition at 50 frames per second” I teach engineering, getting solid understanding on how things work at a low level is key. It lets you then jump up the chain to the top. Starting at the top gives you a tenuous grasp, and you are more likely to fail in the long term.