In recent years, you have heard more and more about “Artificial Intelligence” (AI), “Deep Learning” or “Big Data”. The underlying concepts of these IT disciplines are not always easy to understand but impact your professional daily life. At Moovency, we offer the AI-based KIMEA tool, which improves the analysis of risky postures that can cause musculoskeletal disorders (MSDs) in industries. Our approach, based on a single camera, is about its simplicity of use and its performance. Gone are the tedious placement of twenty sensors on the worker’s body, no more complex training in human anatomy and technology, and no more frequent calibration phases of the system that disrupt the production line. We’ve been betting on AI technologies applied to computer vision for several years now and we’re explaining why.


Computer vision allows your machine to spot humans in images filmed by a camera and determine the positions of each 3D joints. To do this, your computer uses models to deduce these postures from the images you give him Give. These algorithms learn from a lot of data for example, this is called deep learning or “Deep learning” learning.”

During our research activities, prior to the creation of Moovency, we studied computer-based computer vision technology based on the KINECT camera Microsoft. This technology, from the world of video games, allowed to obtain a person’s 3D postures without having to put sensors or sensors on them a combination on the body. Although promising, this technology suffered limits such as the low accuracy of postures when obscuring the field of view by objects. We had then developed algorithms specific to the constraints of the industrial world, we sufficient accuracy for a measure in condition of the real work.

Validated both in the laboratory and on production site, as in our partner Faurecia, the results were published in a scientific journal of reference of the field and available for free here. Although our research has allowed us to use KIMEA in many industrial areas, some usage constraints have yet to be addressed. The main constraint was to place the camera securely and in front of the measured worker, which was not always possible.

That’s why we developed KIMEA 360. These algorithms from “Deep “latest generation of learning” allow you to measure the posture of the your workers from the front as well as from the back. In addition, our solution allows you to to follow the worker with a mobile camera, which addresses the limits to have a fixed point of view. This innovation takes us to the next level important, allowing us to offer you a portable and non-invasive tool compatible the majority of industrial use cases.

example of computer vision

All these innovations are the result of the recent development of AI and in particular the “deep learning” methods applied to computer vision. There are many future developments in the field, and we offer great prospects to make you even easier use and business functionality.


You probably need to know other older technologies as well used to measure workers’ postures and associated risks.

These are inertial sensors (IMUs) that measure their own orientations over time. Although resistant to occultations and beyond the constraints of use related to their use (placement on the body, calibration), IMUs suffer from a lack of robustness and precision of measurement. The main problem with these sensors is the drift of the gyroscope (one of the components of the IMUs), which causes a measurement error that is becoming more and more important over time. To correct this error, it is possible to use the magnetometer. However, the latter is disturbed by the presence of iron in the measurement environment, which renders it inoperative for measurements in the industry.

Far be it from us to denigrate this technology, on the contrary, we use it to specifically measure wrist movements, as rarely visible by the camera (objects, tools). The error in measuring the sensors trade is a major problem to get a valid data because it will help you must recalibrate them every 3-4 minutes during the measurement, which is unacceptable for us.

But all is not lost because we have developed our own sensors resistant to measurement drift problems. This new innovation results will be presented in more detail in the next article, then stay tuned 😉

Pierre Plantard