You can enjoy the view of the beautiful Bottighofen at Lake Constance, but you can also define it. Due to our promotion of training, we are constantly offering bachelor and master theses on modern subjects. This results, among other things, in prototypes, which we can build as a basis for further innovation projects with our customers.
Investment promotes innovation - we are aware of this and therefore it's important for us to continuously promote and train our employees. Flexible working hours, a very good infrastructure, modern working tools and, last but not least, a positive and communicative work environment help to grow together and look beyond the boundaries of the market.
The following section ›Innovation Lab‹, with short essays and exposés of former projects, provides a brief insight into the work that has been done so far.
Facial recognition offers not only the possibility of authentication (identification and verification) of natural people as a biometric procedure, but also the statistical recording of personal characteristics for combination with other data. For example, gender and age can be determined or estimated by the computer or contactless access controls can be implemented. BMT develops and operates a system that uses face detection and facial recognition to mark faces in a live video stream and identify them by displaying additional information. New individuals can be added or authorized by simply filming the face in a few seconds via a web portal or smartphone app. The film is broken down into individual facial images with which a facial classification model can then be trained using machine learning. The facial features are extracted beforehand with a pre-trained neural net. BMT will be happy to advise you on possible applications in your company.
Digital product catalogues play a crucial role in e-commerce. However, these are almost always a compilation of data from a large number of different suppliers. This means that the formats are often inconsistent and have to be merged. To do this, the various source structures must be mapped to a target structure. Nowadays, this work is still predominantly carried out manually. Artificial intelligence with the help of machine learning is very suitable for the automation of scenarios with repetitive processes. BMT has decided to invest in this challenging innovation project. In a first step, the mapping is formalized in order to be able to set up a system capable of learning (active learning). A Mapping-Recommender User Interface (UI) then increasingly provides better suggestions for mappings to speed up the process. Another field of application of catalog AI is the automated extraction of structured product features from unstructured description texts. The data quality of catalogs and thus their economic value often depends decisively on the availability of product characteristics. They are often missing, incomplete or inconsistent. With the help of Natural Language Processing, candidates can be identified for enrichment (feature enhancement). Subsequently, a quality control can be carried out in an Active Learning System. The procedure scales enormously, since in most cases individual features only have to be generally confirmed or rejected once in order to be valid across products or even catalogs. A third usecase of catalog AI that BMT realizes is content-based product recommendations based on similarity of features, description text or product images. Further intelligent applications are planned for the future, which increase the economic value of catalogue data with the help of machine learning, saving a great deal of time.
Image recognition is a field of application of machine vision (computer vision), which has only become practicable in recent years with the rapid development of powerful architecture of neural networks (deep learning) and infrastructure / hardware, i.e. graphics processors (GPUs). If a large number of individual images per object (e.g. different car types) are available, they can be differentiated (classified) sufficiently well. However, the case is problematic in which there are only a few images (often only one) per object, but a large number of different classes are to be differentiated. This scenario is often found with product images in a catalog. If the user of a webshop app is to be enabled to recognize product images with the help of a smartphone camera, more sophisticated technologies must be used, such as "one-shot learning" (computer recognizes objects that have only been "seen" once before) with Siamese neural networks. Unfortunately, this method is still in the early stages of research and development. A person learns continuously and collects an enormous amount of image data when his eyes are open. Thus his neuronal net (the brain) is well pre-trained to master one-shot learning at least in some cases. The computer lags behind in this ability due to a lack of data and insufficient pre-training. An alternative is to transform the problem into a classification of product categories with subsequent similarity ranking. BMT is developing a system for image sorting using machine learning. Initially, this system performs pre-sorting with image clustering (accumulation of images with similar content after feature extraction by a pre-trained neural network). Subsequently, the classification is improved with minimal user intervention in an Active Learning process after naming the categories. The system suggests uncertain candidates (Uncertainty Sampling), whose correct labeling by humans increases accuracy the most. Finally, a neural network can be trained for classification and deployed in the cloud to be available to the webshop app for product image recognition.
What is the best way to identify and support customer requests via various digital channels?
The challenge lies in determining the customer's concerns (target process) and then guiding the customer smoothly through the process.
In a POC we consider different forms of digital communication like e.g. telephone, context search, chat and language assistants (Google Home & Alexa) with the support of AI can better guide the customer through the processes.
Data volumes explode for different and entirely individual reasons. Storage of sensor data, consolidation of systems and long-term data storage are just a few examples. However, what all have in common is the challenge of processing. With Cloud Bigtable, BigQuery and Cloud Dataflow, the Google Cloud Platform (GCP) provides comprehensive and high-performance technologies for processing BigData. In the concrete example, a solution for catalog integration was created, which makes it possible to compare products and prices across providers. In addition, historical data are kept and taken into account so that a product and supplier-related price development can be comprehended.
Whether or not IoT is interpreted as a different name for machine-to-machine (M2M) or as its evolution, this area will continue its triumphal advance and become established in the near future. After BMT had already gained experience with various vendors, the study focused on the advantages, possibilities and prerequisites for the use of the SAP HANA Cloud Platform and to use this platform for the implementation of an IoT scenario. The IoT and device management part of the platform, as well as the mobility and predictive analytics technologies are used here.
Neural networks and intelligent machines are among the current and ever-growing megatrends. How to use already existing and open solutions to adapt them and apply them to specific requirements was investigated in the "Machine Learning - Google Vision API" innovation project. Objects are detected using the Google Vision API and are compared with a previously defined data base. Applications for the use of this technology are just as versatile as exciting. It would be possible, among other things, to order spare parts orders, which are triggered by a photo. Neither the technician nor the customer need to take care of the product number or description because the correct object can be classified, assigned and ordered via the Vision API.
Logistics processes often have the potential to streamline, speed up and improve. In cooperation with an industrial company and an international logistics service provider, a proof-of-concept with a running prototype was implemented. The focus of the solution lies on the "hands-free" work. Augmented Reality glasses provide the user with all relevant and relevant information. This shortens the process and overall throughput time and increases the quality of the work. The result of the work is a feasibility study with a running prototype and insightful knowledge of the technology.
The digitization of maintenance is a further area where the possibilities of "digitally expanded reality" have been researched and integrated into a solution. The scope of functions here is from the data and information exchange with a back-end system (data on the order such as technical space, equipment, material, date etc.) via the guided maintenance, which can be requested via remote video session support, up to navigation And planning the route.