Video Analytics – Embedded vs Server vs Hybrid

VCA (Video Content Analytics) solutions started more than 5 years ago, and they became a solution to all the problems related to intrusion detection, to freeing the system operators of any responsibility, and to the decrease in operating costs within the market.

However, when these solutions started working on the market, they were described by experts as very premature, because at the beginning they did not solve any of the problems, and to make matters worse, it had worsened in some of the cases. Since the detection, or false alarms were a failure, the operating costs increased, and therefore the responsibility of the operators in terms of searching and focusing on issues.

Most of these bad experiences were caused by a lack of preparedness, and because of the complexity of the solutions, especially in the configuration and implementation, as it is needed that a person resets the system and does proper maintenance for a long time.

Today's solutions are more mature than back then, and today there are excellent solutions on the market that have been created in Europe, Israel, America and even Latin America. The key to a successful deployment is, once more, the preparedness and training of technical and sales teams, as well as the client and the operators.

There are several types and architectures of video analysis solutions, each with its advantages and disadvantages. The most important manufacturers in the market provide end-to-end solutions in the IP devices (encoders or cameras) or video analysis done in the servers. But there are hybrid solutions as well, also called distributed solutions (embedded + server).

Embedded solutions

These are the solutions where the video analysis is performed on the IP camera or encoder. The captured image (raw) is analyzed by the processor of the edge device.

The device also checks whether any of the rules has been broken, and if so, it warns the operator in the video surveillance system.

These solutions are typically provided by the IP camera, and encoder manufacturers.

The end-to-end solutions are the preferred solutions for current video analysis, especially when it comes to projects of hundreds of cameras because they need fewer servers, they don't use up so much of the memory server, or processing, they use a lower bandwidth, and can perform several types of rules, such as people detection, vehicles and objects through virtual lines, appearing or disappearing objects, counting and flow, etc.

However, the processing of IP cameras can simultaneously perform only 2 or 3 rules, and does not allow more complex algorithms, such as the detection of smoke/fire, facial recognition or OCR (optical character recognition), such as license plate recognition or maritime containers.

Server Solutions

In this case, as well as being sent to the recording server, and monitoring stations, images of IP cameras and encoders are sent to the servers of analytics. These servers simultaneously analyze the images from multiple cameras, and different rules will apply. If a situation is detected, the system warns the operators.

These solutions are typically provided by the developers of the monitoring software.

These are the preferred systems for small and medium-size solutions, of up to 32 cameras. End-to-end solutions in the servers allow all possible algorithms simultaneously (facial recognition, license plate recognition, etc.). The camera limit, and the algorithms only depend on the server configuration (processing and RAM), and this may be something that can play against them, because perhaps a configuration server can analyze only 16 cameras. This means that for solutions of one hundred cameras, you will have dozens of servers.

Moreover, it also increases the amount of bandwidth consumed as the camera images are analyzed on the servers.

Another disadvantage is that the analysis is not performed on the raw video, but after the compression and decompression process, i.e. when there may be losses and noise in the video. This can lead to false alarms.

Hybrid solutions

Hybrid solutions are the big trend in the market. In this case, IP cameras and encoders perform video analysis over raw images, and send metadata (information and results of the analysis in real time) to a server, which will indicate if the objects, people or vehicles broke some of the rules.

These solutions avoid bandwidth consumption (used only for metadata), allows the use of all algorithms and it is even unlimited. A hybrid video analysis server can support hundreds of cameras, because processing and memory are not used for video analysis.

This type of solution is offered by the manufacturers of video monitoring software, and by some companies that are involved in the process of making video analysis.

Besides all these advantages, the metadata storage allows forensic video search. We will explain this below.

The only downside of the hybrid solution is that it does not recognize license plates, containers, and face recognition.

Forensic video search

The storing of metadata in databases, which is generated in the hybrid solution, allows the search of events and situations that happened in the past.

The video search can be done through different types of filters, such as:

- Target Type: person, vehicle, object

- Type of Situation: appears, disappears, counting, virtual line, etc.

- Color of what is sought

- Time

- speed

- Size

- Image Area

Therefore, it is possible to find, for instance, a small red vehicle at a specific day, and time. Or an adult wearing a green shirt who is walking on one side of the sidewalk.

The main purpose of the forensic search is to reduce the time required for the search of a situation. This solution is essential in systems with hundreds of cameras.

In summary

We made a chart comparing the main characteristics of the three types of solutions:


Despite being a big trend in the market, hybrid solutions will not replace other types of systems. This is due to the fact that they reach different markets and purposes, and they also do not have the capability of recognizing some rules that the market needs, such as license plate recognition and facial recognition.

Still, hybrid solutions are the most recommended for large systems, the benefits in reducing the use of bandwidth and servers, as well as the forensics video search. The three types of solutions are completed in a project with multiple applications.