NtechLab is widely known in Russia and throughout the world as the creator of the most accurate algorithm for face recognition. However, you might not know that our research lab is constantly hard at work on developments of artificial intelligence applications in related fields.
Recently, the company announced a product with silhouette recognition and movement tracking capabilities. The distinctive feature of the algorithm is that it not only identifies a person’s silhouette, but also tracks his further movements even in the crowd of other people within the reach of CCTV cameras.
Each person has a unique silhouette, but it’s not persistent. Many different factors influence the formation of the features vector: height, size, clothing and others. As early as the product development stage, the algorithm recognizes silhouettes with great accuracy. Currently we aim at recognition accuracy of 80−90%, and in the process of finalizing we plan on bringing it to 97−98%.
It is essential to consider that the silhouette “lives” for a limited time: the vector of features will be different if you change clothes, put on a hat, glasses or take an object in your hands.
How to identify a person by a silhouette
To be certain that the silhouette belongs to a specific person, you need to have his face caught on camera as well. This can occur either at the beginning of the path or at any other point. Identifying a person by using silhouettes is an effective, yet not the only application for the new technology. Let’s look at cases where it may be indispensable.
Combining silhouette and face recognition algorithms
For instance, a bicycle has been stolen in the downtown area. The camera is located at a considerable distance: it did not capture the face, but recorded the silhouette. So now we can track how far a person has traveled with the bike, create a route and trace it to a point where there is a camera that can recognize the face.
Inter-camera tracking allows us to track a person’s route and trajectory using CCTV cameras at a single site or in a whole city.
Silhouette recognition as a video analytics tool
The silhouette recognition technology can also be used as a standalone module. In this case, it solves a number of different tasks: It collects data for intelligent video analytics, which are of great value both for public safety and for various business tasks.
We can distinguish the primary tasks that are solved particularly successfully by silhouette recognition:
- Instantaneous accurate tallying in crowded places
- Route detection by silhouette
- More efficient retail spaces
Specific features of the technology
I don’t know you, but I will analyze you.
One of the key features is the impersonality of information about the silhouette owner. On the one hand, it is a drawback: the silhouette cannot serve as evidence of an offence, and it would not be possible to issue a fine for smoking in a public place by a silhouette alone. On the other hand, it has an edge in terms of access to intelligent video analytics where there are restrictions on the processing of personal data. In some cities and countries there are prohibitions on the use of facial recognition. The use of silhouette recognition would be an ideal solution in such cases.
Only a rough visual? That’ll do!
Another important factor for practical application of the technology is the reduced requirements for the video cameras quality. Essentially, this means that ordinary city surveillance cameras will be suitable for silhouette recognition, as long as they are mounted high enough. It is a challenge for developers to teach the algorithm to work with different video streams, including color and black and white, with different shooting angles, and to identify the silhouette of the same people from these streams. We are now steadily accomplishing this task. There are no special requirements for the quality and clarity of video material, they are lower than for footage used in face recognition.
Smart video analytics based on silhouette recognition
Identification of silhouettes for the life of a megapolis
Analytics of group movements in the megapolis allows to optimize the logistics of passenger flows. Simply put, a citizen won’t have to wait 20 minutes for the bus to arrive at the bus stop. There won’t be excessively stuffed or empty buses on the route, and should circumstances change, the city administration will be able to make adjustments virtually in real time.
NtechLab successfully used face recognition video analytics to optimize logistics during the World Cup in Moscow. With silhouette recognition it is even easier to track and count a large number of people in a public place.
The technology has also proven to be effective in enhancing security at public events, searching for missing people or violators and identifying accomplices.
Silhouette Recognition for Business Tasks
The scope of the algorithm application in business is almost limitless. It is always important to understand the movement patterns of people and to benefit from this knowledge.
For example, knowing the specifics of customers’ movement in the mall, the landlord can regulate the rent, and the store owner can consider the additional entrance design or modify the arrangement of goods inside.
Travel companies will be able to plan their excursion routes better by understanding where the city’s visitors are going from the airport.
If your business is targeted at a wide audience that can be localized in a specific place, contact Customer Service to discuss the implementation of silhouette recognition for improved business performance. Your company can be one of the world’s first to use such a precise and effective tool of sales increase. The case itself will be of interest to many mass media, and we will be sure to cover it.