In short, computer vision offers the possibility to make images such as photos and video streams visible to a computer. Give a computer eyes!
For example, you can give photos a name so that they can be found, or to search for photos that are similar. But also to establish similarities and to look for new relevant photos. Google uses this, for example, with google photos to find people in your albums. For a computer, video is a sequence of photos. Kind of a photo album that you can view per photo as a whole (subject, story line). In this way you can analyze and compare moving images.
That is useful for facebook and google. But there are many business applications for computer vision. In industry, a lot of measurements are done with sensors, but this can be expanded with images. For example, you can count products on a conveyor belt. Or think of safety checks that people are not allowed to come too close to dangerous things. You can therefore use our flagship Intra perfectly in the industry.
In the security sector you can use computer vision to distinguish things from people. You can then establish a relationship in the interaction between people and things. But you can also analyze behavior, such as aggression, theft, running away or leaving luggage behind.
A frequently used application is person recognition, which can be used, among other things, for access control at companies. It is also used to access websites or applications. We focus on anonymous applications, without facial recognition.
Computer vision is classified under Artificial Intelligence. Of course you can use machine learning very well with computer vision. This is one of Centillien’s distinctive capabilities. We are active in both disciplines and look at what fits best or whether the combination is necessary. The video below shows that you can also measure distance between people.
As high reliability and stability of recognition is necessary, machine learning will increasingly be a necessity. But if there is a considerable margin of error, this can be omitted. If something does not have a fixed mathematical shape such as a circle or a square and you want to recognize it uniquely, there is only one option.
Our Intra software compares (video) images with each other and a match is made on the basis of statistical commonality. The more images, the better the system works. Usually, Intra has to be trained for a specific task. This is done much faster with images than with normal data, where mountains of data are often required to recognize patterns. Images contain a relatively large amount of data, which makes training faster.
After the recognition phase you can create rules about what needs to be done next. That could be a counting, a distance measurement, temperature or something much more complicated as behavior. If something interesting is recognized, you can send a message or control a robot. Analyzing and recognizing images is quite computer intensive, we have improved this drastically by using graphics cards for processing the algorithms and the image analysis. Performance is an important topic to include in any computer vision project. By adding our own algorithms, we are currently a frontrunner in the field of performance and we are investigating even faster and cheaper ways.