We have expertises in different domains, we list here some of the main areas where we can help
Automatic image processing allows a computer system to manage processing on an image or video stream quickly and without human intervention. These processes make it possible to transform information in the form of pixels into more manageable information such as text information or even easy-to-manage json variables.
We mainly master OpenCV as an image manipulation library. We can also use some components from other known libraries such as DLib that have shown an exceptional performance advantage.
For other specific needs in Deep Learning, we use the Pytorch framework to train a detection model.
Image processing requires in most cases calibration and adaptation to the equipment used. Stereo or 3D vision depends on the type of sensor and its characteristics. The processing itself sometimes requires graphics cards or hardware acceleration modules, especially in the case of embedded systems.
Since 2010, we have been monitoring the various technological advances in terms of capture and low-power hardware acceleration. We can advise you which sensors and cards to use according to your needs.
Deep learning-based methods have existed for several decades. It was only in 2012 that these methods became the most important ones when a convolutional neural network technique won the ImageNet classification competition reducing the error rate from 25% to 16% all at once (contrary to iterative evolutions of ~1%).
Every year, at least one major evolution in the form of a new architecture or mechanism pushes the limits of deep learning. We are following all these major developments and we are trying these techniques as soon as possible. We can help you solve image processing problems using these new templates.
There are several learning techniques that are very effective on a small data set. Random forest or support vector machines (SVM) have shown their performance in solving a large number of problems. Some techniques even demonstrated results in 2019 that exceeded the performance of other Deep Learning techniques. We help you to use these techniques, which are sometimes part of a large treatment cycle.
By defining the context well, many problems can be solved directly with techniques that do not require learning (Machine/Deep Learning). Mathematical or heuristic algorithms can be used alone, or in combination with learning techniques to solve a problem. Thanks to our experience, we can help you optimize your treatments in terms of accuracy and speed of execution.
Since 2010, we have been working on the most widely used video codec standards such as h.264 and those that do not require royalties for use (vp8, vp9,…).
If your job is to manipulate video, we offer you advice in the construction of the multimedia pipeline (audio/video) and the transformation modules. The most popular codecs generally have pre-existing acceleration modules in the hardware, which allows you to manage a large flow on low-resource hardware.
We mainly use the GStreamer framework for our multimedia manipulations. We can develop modules specific to your needs such as supporting a new frame grabber or camera. We can also develop image processing plugins (facial detection, object recognition, transformation,…) on these flows while operating in real time.
Even if the daily development tasks are important in the growth of a company, scientific research guarantees its future. Nothing is more destructive than being technologically outdated or missing an important development announced in a scientific conference. We follow very closely the announcements and articles published in the major scientific conferences. We also carry out internal research and development work to test, adapt and develop these developments to our needs and the needs of our customers.
We offer consulting services if you have an idea of a project and want to know the latest news in terms of state of the art. Starting with an analysis of your real needs, we write a document that will contain the latest advances in the fields related to your project and our own recommendation of choices to use.
We conduct internal research work related to a subject of gesture recognition on and above a table. This work covers the fields of image processing, Human-Computer Interaction, and Software Engineering. The latest algorithms in artificial intelligence and image processing are applied, which also opens up new possibilities for new HCI techniques.