Industrial Image Recognition ML Engineer -- 3
Budget: ₹12,500 – ₹37,500 INR
I'm seeking an expert Machine Learning engineer with focus and experience in building self-healing Image Recognition ML models. The model should be robust and optimized enough to:
- Recognize objects in images, largely taken in an industrial setting
- Effectively address any issues by self-learning and mending itself
- Efficiently adapt to any changes in the imaging environment
Skills and experience for an ideal fit would include:
- Proven experience with Image Recognition ML models
- Understanding of ML algorithms pertinent to industrial environments
- Ability to design, implement, and enhance self-healing features in models
- Exceptional problem-solving skills
Current Issue:-
So we have our own automation and for STB we develop our own software.
It’s basically a server with a capture card that captures live video from STB.
So each server has 4 capture cards as shown in the figure STB1 to STB4
So what we do is we have test cases written and we do automation through remote keys. So usually what we do is we press a key and an action is performed. For example, we have to go to settings so we press the xfinity button it goes to the settings icon on the screen and for example we have to enable CC (close captioning) so we will go preferences and what it will do is we have a functionality where we can track we can do OCR and other stuff on it.
So what we do is we create a region fox example I have created a region here in this picture “Accessibility Settings” and I’ll save it in db and then will do the OCR on it (Optical Corrector Reader). So we are using a library called Tesseract so we pass it to the image and it identifies what is written on screen in that image just like how number plates are recorded in the CCTV’s.
So on the test cases on STB there are combination of doing key passes on the remote. So we have a IR system, we are sending IR signals to the STB and it’s doing the action. So for each action what the output is we verify it with the region.
So sometimes what happens the test fails. So suppose for example we have this guide and I created the region over here (as shown in the screenshot). So it moves any of the word (Press, Guide, to change view) to a little bit up or down. And the test will fail and say I cannot see Press Guide to change view because the location has moved.
So what we want is we want auto healing and it should be able to detect okay this one is failing so how by using machine learning we can improve the accuracy of the test.
So we have lot of regions, images that we compare. For your reference a region means the X,Y coordinates on the screen.
- Recognize objects in images, largely taken in an industrial setting
- Effectively address any issues by self-learning and mending itself
- Efficiently adapt to any changes in the imaging environment
Skills and experience for an ideal fit would include:
- Proven experience with Image Recognition ML models
- Understanding of ML algorithms pertinent to industrial environments
- Ability to design, implement, and enhance self-healing features in models
- Exceptional problem-solving skills
Current Issue:-
So we have our own automation and for STB we develop our own software.
It’s basically a server with a capture card that captures live video from STB.
So each server has 4 capture cards as shown in the figure STB1 to STB4
So what we do is we have test cases written and we do automation through remote keys. So usually what we do is we press a key and an action is performed. For example, we have to go to settings so we press the xfinity button it goes to the settings icon on the screen and for example we have to enable CC (close captioning) so we will go preferences and what it will do is we have a functionality where we can track we can do OCR and other stuff on it.
So what we do is we create a region fox example I have created a region here in this picture “Accessibility Settings” and I’ll save it in db and then will do the OCR on it (Optical Corrector Reader). So we are using a library called Tesseract so we pass it to the image and it identifies what is written on screen in that image just like how number plates are recorded in the CCTV’s.
So on the test cases on STB there are combination of doing key passes on the remote. So we have a IR system, we are sending IR signals to the STB and it’s doing the action. So for each action what the output is we verify it with the region.
So sometimes what happens the test fails. So suppose for example we have this guide and I created the region over here (as shown in the screenshot). So it moves any of the word (Press, Guide, to change view) to a little bit up or down. And the test will fail and say I cannot see Press Guide to change view because the location has moved.
So what we want is we want auto healing and it should be able to detect okay this one is failing so how by using machine learning we can improve the accuracy of the test.
So we have lot of regions, images that we compare. For your reference a region means the X,Y coordinates on the screen.