Geek Talk about pose recognition
The following post was written by Paul Hodgson the engineer behind the data analysis and pose recognition of SmartMat. Warning: Geek talk ahead
So, how is this actually going to work? How will SmartMat know which pose you’re attempting, and how to guide you? What makes it so “Smart”? I’ll try to answer these questions by giving a view from 30,000 feet, and despite being a geek, I’ll try not to make it too “geeky”!
We’re researching a particular field of Computer Science called “Computer Vision”, or CV for short. This is a theory which, as the name suggests, studies (at the most basic level) how Computers can be used to visualise the world around us. An extremely simplified breakdown of the theory is that Computer Vision tries to mimic Human Vision, in such a way that CV can be applied to solve real world visually based problems using computers. The reality is, however, that CV is an extremely complex and sophisticated field of study, made even more so by the fact that computers obviously don’t have eyes and optic nerves, and “see” in a stream of 0s and 1s.
There are already many examples of real world problems being solved with CV. Unless you’re a cave dwelling hermit, the chances are you unknowingly undertake a dozen activities every day in which CV has played a role, whether directly or indirectly. For example, when you upload a photo to Facebook, CV is used to recognise the presence faces in those photos in order to give you the opportunity to “tag” people. Have you ever wondered why chocolates are always in the correct spaces in their boxes, and every box has the same layout? CV theory is applied to create an algorithm which tells the robotic arms in factories used to pack boxes of chocolates, how to recognise a particular chocolate and therefore put it in the correct compartment – this mimics a human recognising a chocolate, and correctly placing it. CV can be used to recognise License Plates on cars to automatically process speeding fines. It’s used to spot flaws on PCB (Printed Circuit Board) production lines (PCBs are in every piece of electronic equipment you own). If your digital camera auto-focuses on faces when taking a picture, then CV theory has been applied in programming the camera’s firmware (the brains of the camera) to tell it how to recognise faces within the frame and subsequently automatically focus the lens on them.
Pretty much anything that involves a computer recognising something visually, and doing something based on the result has its roots in CV theory.
SmartMat utilises the same concepts to solve the problem of pose recognition. Data sent from the Mat to the SmartMat Mobile App will be translated into a format which can be visualised by the computer (in this case, the smartphone/tablet), enabling it to determine which pose you are currently in. Once this has been established, the app will then cross reference the data sent from the mat for that pose with data for the “perfect” pose contained in a database (the data used to cross reference will be dependent on your body shape and size/height and derived from a multitude of Yoga professionals of varying shapes and sizes), thus enabling the app to give visual and audible feedback regarding the pose.
What’s more, if you wish (and consent), your “perfect” pose data can be submitted to the SmartMat server in order to train a bespoke pose recognition algorithm, making the SmartMat adaptive, and continuously learn your style.
Exciting stuff, eh?