Beyond Buzz

AI advancements and abundant video present real opportunity

• By Quang Trinh
• Mar 01, 2022

Artificial Intelligence (AI), Machine Learning, Deep Learning, and Computer Vision are the latest buzzwords spreading throughout the security industry. As with many buzzwords and trends of the past--like the cloud and cybersecurity--they can be intimidating to grasp through all the hype, misinformation and sometimes even negative depictions.

When it comes to the latest wave of popularity around AI and all its various incarnations, how will the security industry respond? The answer lies with how this toolset will be adopted by vendors, system integrators and end customers.

First, it’s important to realize that AI and its subsets are tools and not solutions to security challenges in and of themselves. However, these tools have become ever more powerful as AI has evolved over the last decade.

The evolution of Machine Learning and Deep Learning has been fueled by data, and not just structured data from a database, but unstructured data from images and video as the main datasets (i.e. Computer Vision). Consequently, these tools have become very useful to the security industry given the prevalent use of cameras and video.

The move from Machine Learning to Deep Learning and Computer Vision
With the advent of the internet, the world shifted to become a more connected, digital-rich place. The security industry recognized the opportunity that this presented and it took its first step into that digital world several decades ago by transitioning from analog to IP video. This transition has now opened the gates to a vast amount of a new data including images and videos.

Early on, the internet and the structured data that it provides, lent themselves to advances in Machine Learning which is designed to identify and recognized patterns. This is precisely why Machine Learning has been used in algorithms for e-commerce for several years.

When it comes to the unstructured data provided by IP cameras, recent advancements in Deep Learning techniques (algorithms based on simulated neural networks intended to mimic the functioning of the brain) and Computer Vision (algorithms intended to interpret visual information and derive meaningful insight) allow us to make better use of images and video. For example, these visual assets provide perfect datasets for Deep Learning by allowing us to build excellent models for image classifications. Clearly there’s a synergy (and tremendous opportunity) that exists between the security industry and the AI community.

Think about it. With abundant digital images and video from IP cameras, the security industry is poised to fuel the data demands of the AI community, and in turn, advance its own initiatives. This collaboration has profound implications beyond public safety and security because it offers us the ability to tackle new end customer challenges. For those who are able to recognize and take advantage of opportunity, AI, Machine Learning and Deep Learning present limitless possibilities.

A future built on a solid foundation and current successes
Already in the last few years, many end customers have begun leveraging the network, their IP cameras and video analytics to gain operational efficiencies as well as generate marketing and sales insights. And there’s every indication that this use of AI, and the demand for business intelligence, will only accelerate over the next decade.

What’s more, as end customers become more familiar with AI-enabled technology and realize its value, technologies like Machine Learning and Deep Learning will continue to become less intimidating. In fact, the AI community has actually spent years building a solid engine for Machine Learning through open-source projects and existing frameworks for developers to leverage. Now the critical piece is to provide enough data that will allow a model to learn, refine itself and optimize that data through reinforced and supervised training.

What does this mean for the security industry? It means improved results through greater accuracy and faster response like improved video motion detection and object classification. In fact, some of the current AI-based analytics are better than ever at classifying people and vehicles. Reduction in false positives have always been a critical pain point for the industry and a shortcoming of many analytics, but AI-based techniques in Machine Learning and Deep Learning will reduce or eliminate them completely.

The importance of data quantity, quality and supervision
In the development community there is a term called GIGO (garbage in, garbage out). This concept applies to Machine Learning and it is important to stress that, while quantity of datasets is essential to build a model, the quality of the data is key to the performance of that Machine Learning model. The data from images and video begins with the edge devices and their sensors. Unfortunately, sensors and edge devices are not apples to apples. The Computer Vision data that a sensor provides and processes at the edge relies on the quality of the sensor and the chipset that’s processing that raw data into an image or video.

It bears reminding that computers can’t distinguish color like a human eye does. Computers use numbers--1s and 0s--so color from a computer perspective is represented in intensity through a number scale known as RGB (red, green, blue) from 0 to 255. This means that every device will have slightly different readings of an object’s color in real life.

A user might get one reading that indicates a rich red color, but another device might interpret that red color as a lighter hue. Accordingly, environments with dynamic lighting or harsh shadows can be cumbersome in the overall performance of an analytic.

Machine Learning will output what it has learning and subtle changes in color can impact the performance of a Machine Learning model. However, AI-based analytics have the ability to improve accuracy and performance through retraining of the model with new and revised datasets. This particular approach is a critical differentiator between static analytics and AI-based analytics because customers benefit from improvements from AI-based analytics over the course of time. Accordingly, AI-based analytics can adapt to their environment--which is quite remarkable.

AI advancement through openness, collaboration and education
AI, Machine Learning, Deep Learning, and Computer Vision are transformative technologies, and future innovation will come from a collaborative effort by many in the industry.

Accordingly, open-source communities and their open platforms and frameworks are accelerating innovation and increasing accessibility. Any value gleaned from restrictive, proprietary technology is fading. For system integrators, AI adoption will be fueled by the end customer use cases, and it will take a collaborative effort from the system integrator, vendors, and end customer to develop solutions that work. Education and training around AI technology will also be essential in the next few years in order for system integrators to remain competitive and meet customer needs. What’s more, the ability for AI-based technology to solve new customer challenges beyond just safety and security, will bring new business opportunities.

In the near future, AI and its various subsets will cease to be buzzwords. They’ll become more commonly understood, available and essential tools as the security industry continues innovating for a smarter, safer world.

This article originally appeared in the March 2022 issue of Security Today.