Video Analytics 101

Florian Matusek talks to Mats Thulin, Director Core Technologies at Axis Communications. In this conversation, they talk about how AI, and specifically generative AI such as Large Language Models, is changing the security industry. They talk also talk about the responsibility of security manufacturers to use AI ethically and make sure AI models are not only accurate but also follow responsible AI principles.

Florian Matusek talks to Cees Snoek, professor of computer science at the University of Amsterdam where he heads the Video & Image Sense Lab. In this conversation, Cees talk about how he focuses his research on real-world applications and working together with other scientific fields to solve problems that couldn't be tackled before. He also shares his views on industry collaborating with research instutions and how University spin-offs work in practise.

Florian Matusek talks to Dr. Jerome Pasquero, Principle Product Manager at Sama. In this conversation, they talk about data labelling, synthetic data, and generative AI such as ChatGPT and LLMs in general. They talk about what the future holds for data labelling and how to upskill workers so they are not left behind when more and more synthetic data is powering machine learning models.

Florian Matusek talks to Prof. Matthias Uhl, professor for the societal implications and ethical aspects of AI at the Technical University Ingolstadt, Germany. In this conversation, Matthias shares results of his research in ethics and AI, how he utilizes folk intuitions to get insights from the people actually affected by AI systems and why considering moral questions on implementing AI systems goes beyond regulations and checklists.

Florian Matusek talks to Jean-François Gagné, former CEO of Element AI, which he founded together with Yoshua Bengio and later sold to ServiceNow. JF also helped draft the AI Act of the European Commission and is a proponent of AI regulation. In this conversation, JF talks about the history of Element AI, how he got to work with Yoshua, AI risks and opportunities and where we go from here.

Florian Matusek talks to Marie Rosenkrantz Lindegaard, a professor of sociology at the University of Amsterdam and research leader of the group ‘crime in context’ at the Netherlands Institute for Crime and Law Enforcement. Her research focuses on situational aspects of violence and conflict with particular focus on real-life interactions as they unfold in time and space. In this conversation, Marie talks about how combining video analytics with sociology helped the Dutch police during the Covid pandemic and how this combination of fields has the potential to solve the replication crisis.

Prof. Dr. Y. Raymond Fu is a widely-renowned AI expert, Professor, and a successful serial entrepreneur. He is Distinguished Professor affiliated with both the College of Engineering and the Khoury College of Computer Sciences at Northeastern University. In this conversation, Raymond shares his insights into the current state of AI, the relevance of edge computing for AI and what he learned starting multiple businesses next to his career in academia.

Is it possible to count the number of passengers in a vehicle? Or to get a warning if the driver of the vehicle is distracted? Florian Seitner, CEO of emotion3D, shares with us how to increase safety in vehicles with video analytics.

When it comes to video analytics, we hear a lot of different terms being used interchangeably: Artificial Intelligence (AI), Computer Vision (CV), Machine Learning (ML) and Deep Learning (DL). In this episode we unpack all of these and explain how they are not the same but do relate to one another.

In this episode of Video Analytics 101 Florian Matusek talks with Paule Brodeur, Manager of the Computer Team of Genetec’s AutoVu line of products, about how license plate recognition works technically and what the future holds.

There is a lot of talk about artificial intelligence, machine learning and deep learning. But how does it actually work under the hood? In this episode, we explain how to train a deep learning network and the differences between training and inferencing and classification and detection.

The UK-based start-up Liopa is working on a unique technology to automatically read lips in video, with potential applications in the security industry. But how ready is it for prime-time? In this episode of Video Analytics 101, Florian Matusek is talking to the CTO of Liopa, Fabian Campbell-West, about how it works and where we go from here.

Face recognition is a topic of much interest and controversy. On the one hand, we use it every day to unlock our phone or at passport control at the airport. On the other hand, we hear of instances where it doesn't work and causes more harm than good. In this episode, we explain how every face recognition system works under the hood and why bias in face recognition is of great concern.

In this episode of Video Analytics 101, Florian Matusek is talking with Asaf Birenzvieg, CEO and co-founder of viisights, to discuss how weapons detection using video analytics works, how behavior recognition can help increase accuracy, and what is currently possible and impossible with this technology.

In this episode of Video Analytics 101, Florian Matusek is talking with Nick Ni, Sr. Director for Data Center AI and Compute Markets at AMD to discuss the issue of Dark AI Silicon, and how GPUs for video analytics processing get less and less efficient and what to do about it.

The number one question customers like to ask video analytics vendors is: how accurate is it? And then again, nothing is more complicated to calculate. In this video, we will introduce the F1 score, an established and standardized way to calculate video analytics accuracy.  

The material used in this video is copyright of the Wikipedia user Walber. To learn more about the F1 score, check out the Wikipedia article about it: https://en.wikipedia.org/wiki/F-score

Video surveillance cameras are typically being used for security purposes. Increasingly, however, automation and security begin to merge and protocols such as MQTT provide a standard way for cameras to communicate with other IoT devices. In this session, we are talking with Philippe Henaine of i-PRO on how cameras are becoming IoT sensors and what new applications this evolution brings.

Nowadays, it feels like everyone is an expert in AI and video analytics. But there are a lot of terms floating around that actually do not mean anything and should not be found in any product sheet. In this video we are debunking the top 3 myths: detecting unusual behavior, object tracking and "we use AI", and discuss what we should talk about instead.

Smoke & fire detection has been around for some time. For some vendors it remains just a tradeshow feature but some specialize in this topic, such as Bosch Security & Safety Systems. In this episode of Video Analytics 101, Florian Matusek welcomes Theresa Grunewald, responsible for global business development of the AVIOTEC line of cameras at Bosch.

Video analytics has been used for a wide range of security applications. But one application that is difficult to realize and hence hasn’t seen a lot of attention is violence detection. In this session, we are talking with Gerwin van der Lugt, CTO of Oddity.AI who are specialized in detecting fights and other kinds of violence. We will talk about how the technology works, where to use it, and, more importantly, where not to use it.

Counting persons in video surveillance is becoming more and more popular for a variety of business intelligence and safety applications. But up until now, it was only possible to count up to 30 or 40 people at once in an image. A new type of people counting applications promises to change that and to count hundreds and even thousands of people at once. In this video, we explain how crowd estimation work technically.

Virtually any AI used in the security industry today relies on datasets. However, there are two big issues around datasets that everyone is struggling with. And the second one, licensing, is being ignored by everybody. In this episode we discuss these issues and also what they have to do with broccoli.

2021 marked the release of a multitude of AI cameras: video surveillance cameras with a dedicated deep learning chipset that makes advanced AI functionality at the edge possible. But what are the challenges in building such cameras? What can we look forward to in the future?

Florian Matusek welcomes Mats Thulin, Director of Core Technologies at Axis for the inside scoop.

Edge video analytics and Forensic Search are two hot topics at the moment. And the combination of both enables new use cases with a unified experience. In this session, we are this and more with Aaron Saks, Senior Product and Technical Training Manager at Hanwha Techwin and find out how analytics at the edge can be leveraged for Forensic Search.

Wouldn't it be great if a video analytics system does not have to be configured at all? Running it the back, figuring out by itself what is usual and what is unusual? This is what self-learning video analytics are promising. In this session, we are explaining how self-learning video analytics work and why they are not deployed more widely.

In the past, you couldn’t visit a video analytics vendor’s booth without seeing a demo video involving heatmaps. Today, heatmaps are all but gone from video analytics products. The reason for this is that while they are very visual and appealing, they add very little business value for users. Instead, they are being replaced with occupancy management solutions that provide actionable information.

Do you still see requirements for heatmaps in the market? How do you deal with them?

ONVIF Profile M promises to standardize how video analytics metadata is being exchanged between devices. In this session, we are talking with Hans Busch, Chairman of the Technical Committee at ONVIF. How does Profile M work? How will it evolve in the past-paced AI market? And what is the impact on our industry? Find out in this interview session.

The Covid-19 pandemic brought with it a tremendous demand to control the occupancy of spaces. And the technology used to achieve this is often people counting. In this episode we are explaining how people counting devices work under the hood and what to consider when choosing different technology.

Machine learning is becoming a bigger part of our everyday lives and provides a lot of ways to improve operations and security. At the same time, artificial intelligence is becoming more controversial as applications like face recognition are being regulated in many countries. So how intelligent is AI and should we even call it “Artificial Intelligence” at all?

Automation and smart technology have revolutionized our world. We expect technology to deliver accurate data for security and operational insights while maintaining privacy. In this session, Florian Matusek, and Gerard Fígols Sotero, Head of the Security Solutions Business Unit at Panasonic, discuss the importance of video analytics while protecting privacy.

In this session we are looking beyond the marketing hype and talk about how video analytics work under the hood. Most video analytics today follow a 3 step process, one of which is often forgotten. Check out this episode to find out which steps those are!

If you have been in the phyisical security industry for some time, it might seem obvious what video analytics "is". However, it is not what it used to be and the answer is less obvious than what you might think.