Selected for Rijkswaterstaat Challenge

Studio diip has been selected to take part in the Detection Challenge of Rijkswaterstaat (part of government responsible for infrastructure). The challenge is about detecting unsafe situations on and around bridges. Because detecting people and vehicles is one of our expertises, we have entered a proposal for a system that is able to showcase this functionality in a test environment. During the test day at May 13nd several different circumstances will be simulated that possibly lead to unsafe situations. It is then up to the participants systems to timely warn the operators of these events.

Our current detection software has already been tested on various imagery of traffic situations and for the challenge we have developed a simulation model, so we can reenact specific situations on bridges. Up until the test day we will continue developing our system so it is able to handle the specific requirements that Rijkswaterstaat needs.

The Rijkswaterstaat Detection Challenge is part of the Smart Patrol program, that aims to increase security of object observation by using vessels, smart cameras and drones. Together with their activities, knowledge institutes and commercial parties all sorts of new technologies are researched that can make their work smarter and faster.

 

Studio diip presents: Computer vision demo

Computer Vision demo 

Many projects at Studio diip have to do with using deep learning to train a neural network to detect specific objects. Often though, they are built specifically for our customers and we are not able to show the possibilities to the general public. That’s why we decided to spend some time to create an online demo of object recognition. You can use it to test your own image with some standard algorithms that are trained for different kinds of objects. The algorithms are trained on thousands of photos with different themes and you can test them separately on their own pages: Detecting animals, Detecting electronics, Detecting vehicles and Maritime image recognition. The Computer vision demo detects the specific objects in an image and shows where it is located and with how much confidence.

Maritieme beeldherkenning sloopboten aan de haven

Tug boats in a harbor detected with the Computer vision demo for ships.

Training deep learning networks

For the image detection we use software like: Keras and TensorFlow. Keras is an open source library that is written in Python and works well with the software of TensorFlow. The TensorFlow library is suitable for the development and training of programs that make use of deep learning techniques. With the use of this software we can see how well the algorithm works and optimize accordingly. For the demonstration, the objects are trained on the extensive COCO dataset (Common Objects in COntext). For the examples we have used an algorithm that is trained on a thousand photos with different themes all coming from this dataset. This database contains images of the themed objects that are pointed out by humans and show where the object is hiding in the photo. Therefore the learning algorithm can train itself to automatically recognize what the object looks like in other photos It is also possible to use this technique to detect other kinds of objects.

You can test your own image below, to see how the demo works! 



Image recognition with artificial intelligence (AI)

The technology used in the Computer vision demo can be applied in various ways. For example: We can train a network that analyses electronics on an airport or whenever you want to screen for safety issues on big events. The tool can also be helpful in logistics, science, entertainment or on industrial sites. A beginner can use the tool for educational purposes and an advanced scientist or engineer can use it to gather valuable information to automate complex processes. We can use different online and offline images to detect, count or follow objects. Think about security images, satellite images, heath images or even online streaming. We can embed networks like this for use on personal computers or laptops (Windows or Apple) but it also works on a smartphone (iPhone or Android).

Detecting a herd of zebras

Smart image detection on a herd of zebras

 

 

 

 

Patent for identifying 3D objects

Studio diip has worked for 3D printing company Shapeways to develop a 3D scanner that is able to identify objects automatically. The 3D scanner and custom built software allows the system to match a 3D model to a scan of an object. This allows the system to match the object to a CAD model in a database instantly or even detect damage or flaws in the production process.

In the end these developments made it possible for Shapeways to file a US Patent! (US 2018 / 0104898 A1) We are of course proud to have worked on this system, but also to have Studio diip’s Thomas de Wolf, Guust Hilte and Jop van Rooij listed as inventors in the patent.

You can find out more about our services in 3D computer vision at the types of images page or about our 3D scanning system in the projects page.

3D length measurement with a smartphone

One of the projects we spent a lot of time on this year is the development of computer vision software that can make a 3D scan with a smartphone to determine a person’s height. Together with Ferring Pharmaceuticals, Willem Jan Gerver of the Maastricht University Medical Center and app developer YipYip we have been working hard the last few months to make it possible for patients to do length measurements at home. We can now share the first results of this 3D length measurement by showing you how it works in the movie below!

How does it work?

You can make a 3D scan by walking 180 degrees around a person. The images are then sent over a secured line to a high security webserver. The software we developed uses algorithms that can turn these images into 3D points. When the 3D image has been created they are automatically made anonymous and the original images are deleted. With the 3D image our software can take all kinds of measurements. A 3D image created with a phone looks like this:

Goal

The goal of this software is to make it easier for parents of children with a growth deficiency to do an objective length measurement and immediately share it with their pediatrician. At this moment it takes 4 hospital visits a year and some of these visits are purely for a length measurement. Measuring at home with a traditional measuring rod is not accurate enough to get a good idea of growth over a year. With this application both parent and children get more control and feedback about the effect of the treatment and this should reduce travel time and anxiety as well.

Current status

For now in lab testing the results of a 3D length measurement seem to be on par with a traditional stadiometer (measuring rod) when measured by a physician or nurse. But the 3D technology does require the patient to stand relatively still and that there is enough space to move around with the phone. Therefore there is currently being checked if there is a way to make it easier for example by performing the length measurement on normal pictures. This gives less information about the anatomy of the patient, but should be easier to use.

More information

At Studio diip we are working more often with 3D computer vision project lately. In this project we create 3D models by using normal smartphone images, but we can also create them using special 3D cameras, LiDAR systems and other technologies. We also often work on apps for iOS and Android by adding a piece of computer vision software to them for the app developers to use. The actual app and interface itself is something we don’t usually make, we’d rather leave that to specialized companies like YipYip who are very skilled at this. If you’d like to know more about the possibilities of 3D computer vision or image recognition in apps, take a look at our website or contact us to make an appointment and see these things for yourself.