Derive meaningful information from visual inputs
Here at Dynalytix, we develop Computer Vision applications focused on deriving valuable information from visual inputs like images and videos at a scale and speed impossible for humans.
Our experience in this domain allows us to provide our partners with solutions tailored to their specific datasets, scale and performance needs.
Computer Vision Services
Image Recognition
Image recognition is tasked with detecting and identifying people, items, places, writing or otherwise specific features on an image. Vision data like pictures from different manufacturing steps in a production line, for instance, could be used to make real-time decisions based on real-time detections of any quality or process deviation in real-time.
Historical data could also be batch processed to gain insight on the current production procedure to seek continuous improvement.
We can provide vision systems that can be leveraged across multiple fields by leveraging different tools: in addition to the classification models described above, we can use Image Segmentation methods that are used to divide (segment) each image in multiple areas so that each pixel belongs to a known label.
Object Detection
Object detection adds localization to the former family of models by providing exact information regarding the position of multiple objects within the images. The application field for these models is extremely varied and includes, for example:
- Detection of production issues or defects in manufacturing
- Automation of visual inspections of equipment or buildings
- Warehousing automation and inventory assessments
- Characterising different production stages in the pharmaceutical industry
- Crowd counting in public spaces as well as tools to guarantee that a minimum distance is respected when required
- Workplace safety/PPE monitoring tools
Optical character recognition models convert images of typed, handwritten or printed text into machine-encoded text and are used for translation services, licence plate recognition and road-side signage information extraction.
Face Recognition
Our services also include ad-hoc AI systems like Face Recognition models that match known pictures of individuals to new images even when partial occlusion or angle variations are present.
Using distinctive facial features like the distance between the eyes or the shape of the cheekbones, the algorithms produce a condensed representation of the detected face for easy comparison against a database.
Such capability can be used for:
- access control to restricted areas
- assist law enforcement in identifying criminals and missing persons, even among crowds
- identity validation in a purchase process
Video Analytics
Moving from image to Video Analytics, we can leverage additional information to a series of images (frames in a video) by taking into account the features that are relevant across multiple frames.
Video Recognition
Video inputs can provide supplementary information in subsequent images such as a new angle/view of an object or the evolution of a person’s motion. We can translate this extra data into a deeper knowledge and insight through video processing algorithms that are at the core of functionalities like:
- Human Activity Recognition tools can be used to assess what activities a person is carrying out and are used in security systems, sport and health surveillance, medical and disability assistance, gaming and human-computer interaction, retail theft prevention.
- Visual speech recognition
- Facial and micro expressions recognition
Object Tracking
Going further from video classification, Object tracking models leverage the change in object localization of the detected labels over different video frames enabling technologies like
- Autonomous driving, which at its core consists of an “awareness” of the vehicle position compared to others and the road.
- Gaze estimation to quantitatively measure human engagement and intention
- Speed estimation
Computer Vision Use Cases
Image Recognition
- Analysis of satellite, aerial/drone, remote sensing or hyperspectral imaging for land cover detection, environment conservation, livestock monitoring, urban or agricultural development
- Medical diagnosis and prediction via x-ray, EKG and MRI imaging processing
- Safety monitoring systems that analyse body scanner/metal detector imaging in correctional facilities to detect smuggling.
Object Detection
- Detection of production issues or defects in manufacturing
- Automation of visual inspections of equipment or buildings
- Warehousing automation and inventory assessments
- Characterising different production stages in the pharmaceutical industry
- Crowd counting in public spaces as well as tools to guarantee that a minimum distance is respected when required
- Workplace safety/PPE monitoring tools
Face Recognition
Such capability can be used for:
- access control to restricted areas
- assist law enforcement in identifying criminals and missing persons, even among crowds
- identity validation in a purchase process
Video Recognition
- Human Activity Recognition tools can be used to assess what activities a person is carrying out and are used in security systems, sport and health surveillance, medical and disability assistance, gaming and human-computer interaction, retail theft prevention.
- Visual speech recognition
- Facial and micro expressions recognition
Object Tracking
- Autonomous driving, which at its core consists of an “awareness” of the vehicle position compared to others and the road.
- Gaze estimation to quantitatively measure human engagement and intention
- Speed estimation
Our Projects
The Association of Netherlands Municipalities Vereniging van Nederlandse Gemeenten (VNG)
We worked closely with the VNG team, to identify key areas where AI could bring the most value and evaluated the business data needed to train relevant machine learning models. We also helped provide an additional framework for data collection and strategy to use AI solutions.
We build an AI web-based and mobile-app solution based on computer vision and natural language processing models. The goal strategy is to use AI solutions to help policymakers be more data-driven, cost-effective and efficient in investigating subversive or criminal behaviors of businesses in the main cities of the Netherlands.
How we work?
Mapping the problem
Working closely with your company, we will identify the key areas where AI can bring the most value. This involves meetings with all stakeholders and developing a roadmap for action together.
Data exploration and proof of concept
We will evaluate if you have the business data needed to train relevant machine learning models. If required, we identify additional frameworks for data collection in your company.
Proposal for deliverables
Based on our meetings and data analysis, we’ll share with you the possible AI use cases for your company. We will work hand-in-hand to agree on the desired outcome.
Model development
We will build and apply various machine learning models to your business data, to find the best solution. As a result, we will develop algorithms that accomplish the desired goal.
Application deployment
We integrate the machine learning model with an API or front-end product, making it user-friendly and accessible to the end-user.
Maintenance
Any system might require time to time maintenance, and we are happy to support our customers with that.
Speed up your machine learning project with Turnkey AI solutions
Send us a request and get a free consultation with our AI experts to:- Evaluate your AI use case;
- Determine whether AI should be used to solve your issue;
- Find out how to improve the efficiency of data collection and labelling;
- Learn about time and investment estimates.
Frequently asked questions
In general, if a decision or choice needs to be made based on visual medium features, if a human can do it, then a machine can do it. It just becomes a question of how much data is available for training and how much effort is required to build the models necessary to achieve the business goals.
Data collection is most often on the client side as it is connected to the peculiar business problem to be solved we help as much as we can, especially if it is a data source we have worked with before.
Labeling can be on the client side if very specific knowledge is required for labeling or it can be outsourced to a labeling company or us. Hybrid solutions are also available, and in all cases, we provide proper labeling training and guidelines tailored to the computer vision task.
If you imagine the process made up of iterations of the cycle: data collection, labeling, model development + training, testing & evaluation and deployment into the customer solution, usually the first 2 steps take 20-50% of the time (the smaller the project / standard problem, the higher the percentage) while the split between the latter 3 is really dependent of the novelty of the problem being solved, the required performance level and the complexity of deployment. The computer vision project range is from 5 months to multi-year collaborations.
During the project evaluation, we provide a feasibility assessment, and for problems in which we have experience, we can provide a more precise prediction and discuss the forecasted minimum level of performance.
Powered by Dynalytix