IPVM Video Analytics Course
By Donald Maye, Published Mar 22, 2021, 02:56pm EDTWith over 150 attendees, the world’s best Video Analytics course for video surveillance is being held live.
An on-demand version will be offered in May.
The course costs $299. The video overviews the course:
Based on IPVM's continuous, unique testing program of dozens of video analytic products, we show what really works and how it works today.
What the IPVM Video Analytics Course Covers
In 12 live sessions, over 6 weeks, we examine the following topics in-depth:
(1) Fundamentals 1:
We explain the basics of how video analytics work, including basic image analysis, and the 4 core analytics categories used in video surveillance; VMD, Heuristics, Conventional Object Detection, Deep Learning Object Detection. Additionally, we will examine the pros and cons of each category.
(2) Fundamentals 2:
We introduce deep learning neural networks for video analytics, how neural networks are structured, and the most common open-source neural networks used in video surveillance (e.g. YOLO, SqeezeNet, Resnet, etc.). We will also examine datasets for neural network training (e.g. COCO, ImageNet, Pascal2, Wider, Government datasets), and explain the pros and cons of each, and issues related to bias and ethical challenges.
(3) Measuring Accuracy:
We examine how video analytics accuracy is scientifically defined, including ground truths, false/true positive/negatives, and how they relate to real-world video surveillance performance. Additionally, we will explain how manufacturers define accuracy, simplify their accuracy marketing, and how it is often misleading.
(4) Accuracy Problems:
We explain common problems that break video analytics performance including logistical and practical challenges of large face watch lists, which often cause significant false recognitions in facial recognition. We examine how that relates to average precision and what that means for using facial recognition in access control.
(5) Architecture:
We explain the 9 most common video analytic architectures (e.g. All-in-camera, All-in-recorder, cloud, etc.). We examine the pros and cons with each, including common accuracy issues, cost comparison, and network bandwidth requirements.
(6) Hardware:
We explain the 4 most common video analytic hardware devices (CPU, GPU, VPU, SoC), and examine the common strengths and weaknesses in each, including cost, accuracy, and efficiency. We also explain the most common providers (NVIDIA, Intel, Huawei, etc.) and how each approaches video surveillance.
(7) Person / Face / Vehicle:
We explain how the 3 most common video analytics work (Person, Face, and Vehicle) and most common accuracy problems, and the logistics and practical challenges of performance. We also look at specific metrics that most impact face detection (angle of faces, lighting) and the fundamental difference in face detection and facial recognition.
(8)Advanced Objects / Behaviors:
We explain how the most common behavior analytics (e.g. Intrusion, Loitering, Tampering, etc) work, and logistical and practical challenges that cause them to break. We explain how different analytics (VMD, Machine Learning, Deep Learning) impact performance, and what common problems can be avoided (e.g. Poor lighting, the incorrect field of view, etc)
(9) Facial Recognition:
We explain how facial recognition works, how it is trained/programmed, and what logistical and environmental challenges cause accuracy problems (e.g. camera angles, uncooperative subjects, masks, etc). We also examine common neural networks and datasets that are used by video surveillance manufacturers.
(10) LPR / ANPR:
We explain how LPR/ANPR works, and different analytic types (OCR vs Deep Learning), and how they impact performance and efficiency. We also explain the 5 most common issues/challenges to LPR accuracy (speed, angle, weather, lighting, plate designs).
(11) Demographics:
We examine the most common demographics for people (e.g. gender, clothing, hair, glasses, etc.) and vehicles (e.g. type, color, make, model), and the most common problems. We will look at what features or details video analytics detect for making common classifications.
(12) Providers / Market Overview:
We examine the performance of specific video analytic offerings (e.g. Avigilon, Axis, Dahua, Hikvision, and more), and how video analytics are commonly sold in video analytics (in-camera vs software vs appliances vs cloud). We conclude by examining the future impact of video analytics on video surveillance.
Who Should And Should NOT Take This Course
This course is intended for those who want to manage, design, sell, or support video surveillance systems using video analytics. Completing this course enables you to understand how these analytics work, what problems they have, and how to responsibly sell or deploy these analytics.
This is NOT a software development course nor an academic course in computer vision. If you are interested in building and training your own analytics, this is not the right course for you. Consider watching the Stanford Computer Vision course videos.
Course Calendar
Watch Anytime
Additionally, all classes are recorded so you can watch on-demand on-line anytime.
Certification
At the end of classes, you will take a comprehensive final exam. If you pass, you will become IPVMU Video Analytics certified (see list of IPVM Certified Professionals)
Register
The member's price is $299. This covers classes, personal help, and certification.
If you are not already an IPVM member, registration also includes 1 month of IPVM membership. Register Now.
On-Demand Coming Soon
Moreover, starting May 2021, we will also be offering this course on-demand.
Questions
Any questions, please ask in the comments or email us at info@ipvm.com
5 reports cite this report:
Comments (4)
Typo: Mentioned date is 168 rather than 6 April
I'd like to enroll one of my employees, but his English is just so-so. Depending on the platform you use for the course, we could set up automated translated subtitles. So, what platform/tech do you use to run it ?