Generating Training Data with Mobile Devices
Object recognition with the help of neural networks, which are based on the structure of the human brain, is increasingly finding its way into different areas: Industry and logistics, medicine and agriculture, and of course in autonomous driving. For many applications, their faster adaptability and scalability make neural networks more suitable than traditional image recognition algorithms.
But an artificial intelligence must first learn which objects it should recognize. This requires a training data set that contains image information with additional descriptions, for example object name or object class, size or position. It is true that data sets are already freely available for many everyday objects, such as people, animals, and cars. However, when new products are processed in logistics, for example, until now such initial training data had to be generated with a great deal of manual effort.
One alternative is to use mobile devices such as smartphones or tablets. Thanks to their high-resolution cameras and ever-improving computing power, they can record relevant objects on the basis of images and annotate them largely automatically, i.e. provide them with the required additional data. A flexible tool chain that provides various object trackers, segmentation algorithms, and machine learning models, which with mobile annotation makes it possible to react directly to changing situations - such as lighting, object size and type, and weather. In addition, functionality can already be tested during annotation with the help of the smartphone or tablet. In the future, higher-performance object trackers, segmentation algorithms, and more powerful devices could even allow this approach to compete with conventional processes for generating training data. For details, see our whitepaper "How to Quickly and Efficiently Generate Training Data for Neural Networks", which you can download here.