The awesome power
Vision and Tiny Machine Learning
At TRICAS, we specialize in innovative product development. We frequently research new techniques and innovations that we can use to realize customer ideas or to answer customer questions.
For example, we recently researched and developed the miniaturization of smart vision systems. For this project, we integrated a tiny camera into a handheld device and equipped it with an image-recognition model that works independently of external sources.
At the start of the project, it’s essential to capture the right data. A diverse data set is vital in training effective models for image recognition.
By exploring and comparing different software tools, we determined what types of images the models responded strongly to.
The potential of a miniaturized camera system is unlocked through a neural network, a paradigm in Machine Learning inspired by the workings of the brain.
Through deep learning, the technique by which neural networks are trained, we can develop a robust model for image recognition, capable of identifying and classifying objects.
Traditionally, such Machine Learning models run on powerful computers or in the cloud. With TinyML methodologies, however, we can shrink these models and implement them on embedded platforms such as a microcontroller.
Essentially, it’s like giving a small sensor a super-smart brain (for a very specific problem) so that it can think for itself and make decisions, without needing a big computer in the cloud to do so.
The results of this innovation project have led to new functionalities in electronic devices that include the recognition of consumables as well as the identification of the user and environment.
The photo shows our demonstrator that is able to recognize consumables based on (animal) logos.