Evaluating and selecting AI video analytics solutions for your organisation

0
449

As the multi-billion-dollar market for artificial-intelligence-based video analytics continues to grow, so does the number of video analytics solution providers. In Q3 of 2018, Stockholm-based consulting company Memoori identified 128 active companies in the supply chain for AI video analytics.

This list is far from exhaustive, considering how analytics has been gaining interest and becoming mainstream in 2020, with users expecting more accurate alerts based on object detection instead of motion detection, hardware providers developing more powerful but compact chip sets for deployment, and more startup solution providers carving out their niche in the market.

Given so many choices, the question arises as to how a system integrator can evaluate and select the best solution for his customer. Although the criteria vary for each vertical, there are some key metrics to consider across the field:
Open platform
Ease of use
Robustness and reliable performance
Versatility
Good support and integration
Low total cost of ownership

1. Open platform
Open platform allows the user to have complete flexibility, avoid being locked into any particular manufacturer, and utilise the best-of-breed solution available in each category.
In 2019, an IPVM survey shows that 51% of system integrators always prefer an open platform to an end-to-end solution (i.e., all components including camera, VMS, analytics, etc. provided by one manufacturer), and 24% select open platform or end-to-end depending on customer requirements . For analytics, as the users commonly have an existing infrastructure, investing in a technology overhaul would be too expensive.
An open-platform analytics product, i.e., a camera-agnostic, VMS-agnostic, and computer-server-agnostic product, will add value to the existing infrastructure within a reasonable budget. Open platform also makes it easier and more cost-efficient to upgrade each component when necessary.

2. Ease of use
One of the main reasons and goals of applying AI to security is to help the user automate the process of watching hours and hours of surveillance videos, extract useful information and send alerts when necessary. In other words, AI should make it easier for the user to operate the security system. Thus, a good AI video analytics solution must be easy to set up and connect to the existing infrastructure, easy to use on a daily basis, and easy to scale with the expansion of the business. Let us examine each point in more details:
Easy to set up: a turn-key, plug-and-play solution helps save time and money. The system integrator can spend a couple of hours instead of days to help the customer set up. In both 2018 and 2020, the most common reason that integrators cited for choosing a solution is that “it just works” .
Easy to use: an intuitive, no-learning-curve user interface allows the customer to make the solution second-nature, maximize its utility and gets the highest return on investment. The best-case scenario is that everyone in the user’s organisation, e.g., every police officer in a city police department, can use the solution on a daily basis, not limited to a technical staff with rigorous training.
Easy to scale: the solution must be designed to seamlessly scale in different ways: number of cameras (e.g., from a few to a few thousands); deployment locations (e.g., can we access data in our branch office in another city? how about another country?); types of device and deployment (e.g., body-worn cameras, in-vehicle, control center, cloud).

3. Robustness and reliable performance

Traditional VMD (video motion detection) -based analytics have many limitations and false alarms, so AI-based analytics were developed, primarily to identify different objects in the videos with high accuracy.
However, such accuracy must be achievable in different real-life environments. The best solution does not let low lighting, snow and rain, spider crawling in front of the cameras, etc., interfere with human intrusion detection or license plate recognition at night. In the case of temperature detection, users should be able to walk by the system at a normal pace without removing the mask to minimise disruption and maximise worker efficiency.
A more robust solution means less time and resource spent on false alarms.

4. Versatility
A versatile, feature-rich, multi-functionality video analytics is the most effective choice for system integrators in the long term. Not limited to only object detection, AI can be trained to recognise higher levels of details (e.g., faces, age, gender, license plates), track objects (including people and vehicles), and detect certain behaviours (e.g., loitering, theft).
In other words, a more versatile analytics solution can recognise more types and behaviors of objects for more use cases. Most users have certain pain points today and are looking for only one or a few solutions. However, as the organisation grows, new situations and requirements may arise, which call for new detection functions in video analytics. The costs and complexity will add up quickly if each solution has only one function. A few examples:
An LPR camera may be perfect for the need to record all license plates today, but if the police wants to find a black Toyota Prius with “A23” in the plate number, a solution that can detect the plate number, vehicle make and model will save much more time and effort.
Intrusion detection based on the ability to distinguish human from other moving objects (e.g., animals) is only the first step. What if the user needs an alert for people that enter a construction zone without a hard hat and safety vest? The answer is an AI solution that can grow its repertoire.
In the current pandemic, business must adopt temperature screening, distancing detection, occupancy detection, and mask detection; a solution that can provide all four analytics in one platform is clearly more useful than four individual solutions, not to mention whether the solution can be repurposed after the pandemic has been resolved.

5. Good support and good integration
One of the main reasons that system integrators might select an end-to-end solution instead of an open-platform one is technical support: more responsiveness and less finger-pointing. In terms of responsiveness, good technical support is a part of the ease of use, where the system integrator and the user can rest assured that any question can be answer via email or a phone call to the manufacturer.

In terms of having a one-stop-shop solution to reduce finger-pointing, good support means the manufacturer can provide easy integration to 3rd-party systems, which includes API interface support. One example is access control. Video analytics is a great tool to enhance access security (e.g., face recognition to open doors for employees; LPR for parking management; weapon detection linked to automatic locked-down system), but only 24% of video surveillance systems today are integrated with access control.

Two of the main reasons: (1) integration is expensive, and (2) the systems are not compatible. Both hurdles can be overcome if the analytics solution bridges the gap between cameras and access control system via its API. Cost is always a determining factor, especially in the SMB market [vi]. Customers’ expectations are high, and higher-resolution cameras are decreasing in price and increasing in numbers, which means more data to process than ever. A good analytic software solution is not only capable of many functions, its algorithms are efficient enough to fit more into the same server specs, and it does not require expensive cameras to have good accuracy, thereby increasing cost saving for the entire system.

In summary, these criteria help both the system integrator and the end-user save time, money, and effort and get the most out of video analytics in the long run. A high-performance, versatile, turnkey solution is already a reality with today’s technology, and it will only continue to improve, so there is no reason to settle for less.