ML is everywhere in our digitized world. Apart from typical examples such as the detection of spam mails or recommendation systems for search results and product suggestions, ML algorithms are used by banks in the loan approval process and of course in our smartphones for speech and face recognition. ML has also found its way into the automotive industry and is becoming increasingly significant.
The spectrum of ML applications ranges from intelligent production lines and automated quality inspections to driver assistance systems such as lane and traffic sign recognition, as well as an intelligent cockpit that recognizes either the driver or any distractions while driving. At e.solutions, we are carrying out research and development in this area to make driving safer and more pleasant.
Machine learning is already being used in the latest vehicles to capture the environment or to monitor and control the interior. At e.solutions, we focus on what is known as the intelligent cockpit, which ranges from vehicle interaction using gestures or speech to distraction detection and personalization. For instance, settings such as seat position, mirrors or comfort temperature can be linked to individual drivers. Reliable driver recognition using facial features is just one of our many advanced development projects. Our team also develops algorithms for intelligent caller lists or smart seat heating based on individual user preferences.
It goes without saying that these services must be reliable, but they also need to work extremely efficiently and without connection to the cloud. In addition, some applications, such as the intelligent caller list, need to be able to re-learn in order to react appropriately to routine changes. For this purpose, e.solutions has developed the ML framework “esoAI”, which takes into account the various challenges in the automotive environment.
esoAI not only supports conventional and lightweight ML algorithms such as support vector machines, but also deep neural networks (deep learning) or tools for incremental learning. esoAI provides a flexible and intuitive Python interface which, among other things, simplifies model development and validation through its compatibility with scikit-learn. A backend written in C++ enables rapid execution even on production hardware with no dedicated acceleration capabilities.
