Nowadays, computer vision technology is all around us. It ranges from face detection in a camera application to license plate recognition in a parking garage and the scanning of QR codes. But there are far more applications of computer vision or image recognition.

We apply the technology on normal colour cameras, as well as on 3D depth cameras, satellite imagery, and microscopic images. We use it to detect objects, but also to track, count, and identify them. We develop the software and we are familiar with the hardware that is required for the job.

We often use Artificial Intelligence (AI) in our computer vision applications to get the most useful information out of images or video. When we use this information to make decisions (semi) automatically, we speak of computer vision. But we don't just throw AI technology at any potential problem. We're also accomplished in applying more classical image recognition algorithms if that's the best solution for the case at hand.

In 2020, the team of Studio diip joined Pegamento, adding more than 12 years of experience in developing computer vision applications to the company. Our team of specialised computer vision developers is working on innovative projects with vision at the core every day.

Read on to find out whether computer vision is of added value for your organisation!

What is Computer Vision?

We usually define image recognition as the ability to interpret useful information from pixels in images or video. If you then use this information to automatically take action or make decisions, we call it computer vision.

We often make use of Artificial Intelligence in our computer vision applications. We then train deep learning networks that are able to interpret the visual world around us. Using images or video sequences that are annotated with the thing we want to detect, we can train these AI networks to classify or segment objects. The software can then use these detections to react to the things they 'see'.

What a specific network needs to 'see' is defined by the developer and the way the network is trained. But we can also use more classical computer vision, where we define exactly what to process in terms of pixels or pixel groups. Regardless of whether we use more classical computer vision or AI, we aim to empower users and systems to do their best at the task at hand, either by automating processes or by keeping users informed.

Computer vision technology

For every project, we use a wide array of different techniques for computer vision. We've outlined some examples of technologies we use that might be part of a solution for your own application.

Feature matching

An algorithm that looks at specific pixel clusters in a sample image and matches them with a database full of images. Useful for finding out which object is passing by a camera.

Want to know more about feature matching?

Artificial Intelligence

We use deep learning networks to train systems to detect objects and take certain actions.

Want to know more about AI in computer vision?

Structure from Motion

With Structure from Motion, you can make a 3D scan with a smartphone camera.

Want to know more about Structure from Motion?

We build

  • Computer vision systems
  • Desktop applications
  • Server applications
  • Computer vision apps (iOS & Android)
  • Custom plugins or libraries

Our skills

  • Classical computer vision
  • Artificial Intelligence
  • Setting up an API
  • 2D & 3D vision
  • Structured Light 3D scans
  • Time of Flight 3D scans
  • Structure from Motion
  • Camera and optics
  • Performance optimisation

We develop in/with

  • Python, C++, C#, JavaScript and Java
  • OpenCV
  • Point cloud library and Open3D
  • TensorFlow and PyTorch
  • Android Studio and VS code
  • CUDA and NVIDIA Jetson
  • Bitbucket CI/CD and Docker

The power of computer vision at your fingertips

Smartphones nowadays have plenty of computing power, and connection speed has greatly improved because of 5G networks. This has created a lot of new opportunities for real time computer vision applications on mobile devices.

3D height measurement application

We've developed a 3D height measurement application that makes it possible to get an accurate measurement of a person's height using a smartphone. Parents of children with a growth deficiency can take an objective measurement from home that can be shared reliably with their physician.

Computer vision via 5G

Fish on Wheels (FOW) is a demonstration project for a self-driving aquarium on wheels. FOW consists of an RC car, an aquarium, and a construction to mount a smartphone above the tank. The video from the smartphone is streamed  to a server using 5G. The server than analyses the video to detect the position of the fish and turns that into instructions for the RC car. The smartphone then sends these instructions to the car. Because of the minimal delay in connection, this all happens real time.

Possibilities of Computer Vision

There are many ways to apply computer vision. We've already named a few before, but we'll list the main application categories here:
Detect

Detecting specific objects or people in videos or images. For instance suitcases, birds, or windows.

Identify

Recognise the specific type of objects. For instance convertible cars, specific labels, or face recognition.

Track

Follow detected objects over a period of time. Often used to make sure things aren't counted multiple times.

Register

Storing detected information and registering it in a specific way. Such as counting bread on a conveyor belt or storing weight per car on a bridge deck surface.

Computer Vision application areas

Medical computer vision

A growing area for computer vision is healthcare, specifically E-health solutions that enable people to monitor their own health at home. This can, for instance, include applications to diagnose visual symptoms, or to validate a self-test result.

Malaria detection

A healthcare example is an application to detect and identify specific malaria parasites. A trained microscopist can count and identify malaria parasites relatively easy. But it is a time consuming and repetitive business. An application has been developed to read images from a robotised microscope to make it possible to deliver a pre-filtered overview to physicians.

Detecting unsafe traffic situations

How does traffic adapt to a recent modification to the situation? What type of vehicles pass by specific locations? With computer vision and AI, it is possible to detect unique vehicles and other road users. This makes it possible to follow vehicles over multiple cameras, track traffic flows, or count the occurrence of certain situations.

The BCR system

BCR (Bridge Calamity Recognizer) is an intelligent system that supports bridge operators. The system uses multiple cameras and AI to follow objects on the bridge deck. This makes it possible for bridge operators to have a real-time top-down overview of the situation on the bridge. They can also rewind to see where a specific object came from and to where it disappeared. In this way, they can make a well-informed decision about opening the bridge deck. See more about this case for Rijkswaterstaat here.

We envision more possibilities of using a connected camera system to map traffic situations on an online dashboard. For instance, situations on canals and locks, but also at intersections, in public spaces and in logistic centres. The information from this system can also be enriched by using data from other sources, to make it possible to monitor weight, environmental impact, sound, and wear and tear.

Quality control for production

Quality control is an important aspect of every production process to make sure that  source material meets certain demands or that delivered end products are up to the required specifications.

It is possible to position one or multiple cameras at tactical locations in a production process and use them to see deviations in quality using computer vision. This makes it possible to directly take action or to at least have information about the overall quality.

Quality bread

Maintaining the quality of bread being produced and going to customers is essential for modern bakeries. In-line quality control during the production process makes it possible to identify individual bread rolls that do not meet the quality control specifications and reject them. Traditionally, this is a very time-consuming and tedious manual process. At Amstelveld Bakery, we have implemented a system that uses AI networks and a normal camera to automatically reject bread rolls based on size, shape, and appearance. The first deep learning network detects the bread rolls and what type they are. A second network assesses the visual quality. The bread rolls automatically get rejected by sending a signal to a Programmable Logic Controller (PLC) to open individual lanes.

A look in our lab

  • Vegetable recogniser
  • Bridge Card Reader
  • Industrial Waste detection
  • AR Blackjack
  • 3D scanner
Vegetable Recogniser

A system that suggests recipes based on detected vegetables on an interactive tabletop.

  • Real-time vegetable recognition
  • Intuitive way of discovering new recipes in the supermarket
  • Uses an HTML5 canvas connected to C++ computer vision over webSockets

Interested in the possibilities of applying computer vision in your company?

Computer vision can be applied in many markets and segments. In places where imagery is already used, it is easy to apply, and in many cases a camera is an easy and flexible addition. The technology is innovative, but surprisingly simple and affordable to apply.

Is computer vision a match for your company? What kind of reductions or quality improvements are possible within your processes?

Apply for an introduction using the form to find out how computer vision cam make a difference in your company.