Computer Vision? Crystal Clear!

by | Jun 26, 2018 | Tech

computer vision neural network

Artificial intelligence. Illustration of a brain-shaped printed circuit board.

 

In this article, we wrote you a simplified definition of Computer Vision with concrete use cases, in order for you to understand the future of industries.

For few years now, the Artificial Intelligence (AI) generates community interest in various industries. If you have already heard about AI, the expression Computer Vision should not be unknown to you.

Computer vision in few words…

 

Computer vision is a subdivision of AI. This set of methods and technologies enables the automation of specific tasks from an image. As a matter of fact, a machine is capable of detecting, analyzing and interpreting one or more elements of an image or video stream. Therefore, the machine is able to make a decision and perform an action.

 

We will focus on a specific area of computer vision which is image processing and especially image recognition. However, there are other disciplines within Computer Vision such as facial, text or iris recognition.

 

How does it work?

 

In most cases, image recognition is based on Deep Learning (DL).

For those who are not familiar with Deep Learning, it is a set of automatic learning techniques. In fact, DL is based on artificial neural network, similar to the human brain. Namely, a neural network is made up of several successive layers of neurons. In addition, depending on the chosen neural architecture, each of these layers can influence another.

 

neural network layers

 

Beforehand, it is necessary to train the neural network, in order for the algorithm to be able to recognize an image. To do this, we provide to the algorithm a database, that has been manually annotated. We make the annotation according to the type of information we want to extract.

To be effective, the training data set must meet 3 important criteria:

  • Volume. A rich database will improve the algorithm’s performance rate.
  • Diversity. It is necessary to have a wide range of images in order to accustom the algorithm to recognize the required elements in various situations.
  • Proportionality. For example, if you want the algorithm to recognize 3 different categories in your database, it will be important that these 3 categories are proportionally represented for a proper training.

 

Object detection and image classification, but what is the link with Computer Vision?

 

The object detection and the image classification are two notions regularly associated to Computer vision. Indeed, it is two tasks that image recognition can execute.

First of all, object detection consists in searching for a particular element and locating its within an image, by using a “box. Moreover, there is a more elaborate and accurate detection method (to the nearest pixel) called polygon segmentation.

Secondly, the image classification enable to identify to which category an image belongs, according to its composition. In other words, it identifies the main subject of the image. However, it is possible to associate more than one category to an image thanks to the tagging method with a process similar to classification.

 

How can we use Computer Vision?

 

In practice, computer vision assists humans in their daily tasks, makes it easier for them to detect specific elements, behaviors, situations… In addition, it saves a considerable amount of time but also reduces the rate of human errors (even if the machines are not infallible yet!). Besides, innovations such as autonomous cars or connected objects are now possible thanks to computer vision.

As you may have understood, computer vision can be used in a wide range of industries: from pharmaceuticals and medicine to construction or even the automotive industry.

To illustrate our point, here is a concrete example of computer vision application: autonomous vehicles. As a matter of fact, Valeo has developed smart cars with image recognition technology. They are capable of automatically adapting the cabin temperatures to the needs of each passenger. The system analyses passengers according to their gender and the layers of clothing they wear. The ultimate goal is to save energy while optimizing the customer experience.

 

If you are curious to know more about the different possible use cases, discover the ones we have worked on at Deepomatic (smart video surveillance, self-service checkout, detection of dangerous areas…) !