modular AI solutions
AI in a box:
MODULAR, NETWORKED und applicable
AI combined with Imaging
Examples of our Team BILD (Computer Vision)
Road traffic analysis
Filters for abstraction and simplification of reality enable statistics and analyses by focusing on the essentials.
Security through drone defense
Airports, football stadiums or buildings: Airborne threats from drones must be detected and fended off. Due to the technical possibilities, drones represent a real danger, as they can transport poison into stadiums or damage the ability to fly of aircrafts at airports.
- Deficit and inadequate drone detection by the naked eye due to natural limitations such as distance, weather conditions, visibility obstructing trees or clouds.
- The used infrared cameras allow drones to be detected even in darkness and through clouds.
- Passive systems avoid outsmarting by drones in active radar systems.
Visual information is perceived, interpreted and corresponding actions are derived. We developed this self-driving robot during our first hackathon in Hanoi.
- Camera Logitech C270
- Engine RB03
- Computing Device Jetson TX2
- Recognizing and following correct traffic signs
- Recognizing and not following incorrect traffic signs
- Recognizing and following the road course
From automatic doors or hand dryers to intrusion protection by triggering an alarm.
- Commercially available camera for the office
- Differentiation between employees and visitors in the office with simultaneous anonymization of the person
- Alarm for unannounced persons
- Refusal of access for unannounced persons
- Functionality of motion detection even in darkness
ONE solution FOR ALL IMAGE DATA
Whether still or moving image
… all image, video and stream formats can be connected to the AI and be processed.
… Single, studio and multi-perspective images of a dynamic situation, such as when all cameras in a pedestrian zone are connected, can be connected to the AI and processed.
No matter which electronic camera technology you use
Digital cameras, infrared cameras for heat detection or LIDAR for optimum distance measurement can be connected to the AI and processed.
Our AI SOLUTIONS are…
Even more concentrated software by TEAMS WITH FOCUS
Natural Language Processing
Our WORT team takes care of everything to do with the mechanical processing of natural language.
Our BILD team takes care of everything to do with machine vision. With the help of cameras, image-based data can be interpreted and analyzed.
Sensor Data Analytics
Our ZEIT team takes care of the analysis of highly dynamic (sensor) data for the recognition of events.
Face recognition performs the task of identifying or verifying a person by comparing and analyzing patterns based on facial features extracted from face images. It is widely used for security applications – like tracking criminals more easily or protecting data with your own face instead of a username and password, though there is increasing interest in other areas like augmented reality.
Object detection performs the task of finding all objects in an image and drawing bounding boxes around them. It is one of the most common used techniques in a broad range of applications such as surveillance, security and advanced driver assistance systems.
IoT Malware Detection
Malware detection using machine learning can automatically identify malware from normal programs and has become increasingly important these days.
Anomaly detection (or outlier detection) performs the task of identifying rare items, events or observations which raise suspicions by differing significantly from the majority of the data. It is applicable in a variety of domains, such as intrusion detection, fraud detection, IoT malware detection, fault detection, system health monitoring, event detection in sensor networks and detecting ecosystem disturbances.
Recommender systems represent an approach to develop more personalized information systems with the aim of providing people and organizations with the right information at the right time. Recommender systems have gained considerable traction online, for example, music recommendations by Spotify, film recommendations by Netflix or quotation creation in the sales department of companies.
Intrustion Detection System
An intrusion detection system (IDS) is a system that monitors network traffic for suspicious activity and issues alerts when such activity is discovered. The foundation of any intelligent IDS is a robust data set to provide examples from which the computer can learn.
Fraud detection refers to the detection of criminal activities occurring in commercial organizations such as banks, credit card companies, insurance agencies, cell phone companies or stock markets. The malicious users might be the actual customers of the organization or might be posing as a customer (also known as identity theft).
(in Natural Language Processing)
Attention mechanism (in Natural Language Processing) refers to the idea of allowing the decoder to attend to different parts of a source sentence at each step of the output generation, instead of encoding the full sentence into a fixed-length vector. The algorithm learns what to attend to – based on the input sentence and what it has produced so far.
One of the problems that researchers have struggled with is how to link pronouns to antecedents, which resulted in this old joke, mimicking the call and response of a protest march:
What do we want? Natural Language Processing!
When do we want it? – Sorry, when do we want WHAT?
A neural network armed with an attention mechanism can actually understand what “it” is referring to. That is, it knows how to disregard the noise and focus on what’s relevant.
Text Summarization states the main ideas or facts in compact statements from a single or a bunch of documents. Considering the nowadays information overload, text summarization can help a company in data analysis and fast decision-making.
A Question Answering System refers to computer systems with the capability of answering the questions posed by humans in a natural language. These systems first translate sentences into a machine-representation, then try to find supported data in databases to generate valid answers. Question Answering Systems are becoming more and more popular thanks to Siri and Google Home, so-called virtual assistants.
Sentiment analysis, also known as opinion mining or emotion AI, refers to the automated process of identifying opinions within written or spoken language. Companies may use sentiment analysis to understand what customers think about their products and services. Beyond determining simple polarity (positive, neutral, negative), sentiment analysis can extract deeper understanding of the context such as the aspects and the emotions of the opinions.
Information extraction performs the task of automatically finding structured information from unstructured machine-readable documents. Information extraction, at a basic level, identifies named entity mentions such as products, organizations, locations and time expressions. At a deeper level, information extraction can find events, relationships and coreferences hiding deeply in written text.
Machine translation means the software-supported translation of a language A into a language B and vice versa. It consists of both language analysis and language generation. Machine translation enables people who do not speak the same language to work together and facilitates the learning of foreign languages.
Vision-based action recognition performs the task of identifying and inferring human actions or goals of one or more people based upon a series of observations or video clips. Due to its multifaceted nature, it can be immensely useful in very different fields related to human activities such as home-based rehabilitation, security-related applications, healthcare and logistics support.
Instance segmentation performs the task of assigning a label and entity ID to each pixel in an image in order to identify image regions of same instance (like “person 1”, “person 2”). Instance segmentation can be used in many applications where we need to localize each instance in the image like object detection, but at pixel-level accuracy like semantic segmentation.
Semantic segmentation performs the task of understanding an image at pixel level by assigning a label to every pixel in an image in order to identify image regions of same characteristics (like “street”, “person”). Some of the practical applications of semantic segmentation are in medical imaging, autonomous driving and traffic control systems.
Object tracking performs the task of locating a moving object or multiple objects while maintaining a unique ID for each object over time and controlling the camera if necessary. It is often combined with object detection in a variety of applications such as human-computer interaction, surveillance, security, counting systems.
Image classification performs the task of assigning one label to an input image depending on the image content. It is one of the most fundamental problems in Computer Vision that, despite its simplicity, appears in most practical problems.
Optical Character Recognition
Optical character recognition performs the task of converting images of typed, handwritten, or printed text into editable and searchable data. It allows users to save a lot of time and effort in various applications from exploiting text-related content in images to scanning and extracting information in documents.