AI in Video Surveillance

Rachel Smith, Applied Technology Review | Friday, April 23, 2021

Rachel Smith, Applied Technology Review

The video analytic algorithms are fed a comprehensive data set representing an object using DL. This phase is known as training, and it involves the algorithm teaching itself to recognize a specific object type.

FREMONT, CA:-Video analytics in video surveillance utilizes Machine Learning (ML), and Deep Learning (DL) approaches to define, classify, and assess the properties of objects.  When people obtain new information, their brains try to make sense of it by comparing it to similar objects. This comparative approach is similar to what ML and DL algorithms use.

What do ML and DL mean for Video Analytics? Both approaches define programming techniques that enable a system to learn from a collection of data. Human programmers usually preset or correct the attributes of the data a device searches for while using ML. For example, the system may be configured to mark an object as a ‘person’ if it is taller than it is wide, has limbs that move in specific ways, and so on.

The video analytic algorithms are fed a comprehensive data set representing an object using DL. This phase is known as training, and it involves the algorithm teaching itself to recognize a specific object type. Thousands of photographs of people of various genders, clothing types, ethnic backgrounds, images taken from various perspectives, and more are fed into the device.

The algorithm decides which attributes are identical and which are dissimilar and how to weigh the importance of those characteristics. The algorithm may determine that the majority of images have a triangular-shaped object near the upper part of the image, with two darkened oval spots near the bottom, which one might think of as a nose on someone’s face, after analyzing thousands of images. In reality, the algorithm might have found a slew of other traits that one would not have considered.

Before a customer uses the technology, the creators of the software must train it. The method necessitates a significant amount of computational power, far beyond that needed to detect and identify objects in the field. The output is a file that the machine uses to see whether an object detected fits the classification.

Since DL uses a computer to evaluate object characteristics, it has resulted in analytics that can provide much more granular classification. For example, while older methods may be able to detect an individual, DL-based analytics may determine if the person is a man, woman, or infant. It may also detect an individual’s associated features and the form or make of the vehicle. In most cases, AI in video surveillance is trained at the design stage and does not become ‘smarter’ when it is used in the field. On the other hand, DL and ML have this potential and, when used, can produce analytics that learns over time.

Systems that decide what is common in a scene are examples of typical applications. A school corridor, for example, sees a surge of traffic every 45 minutes or so between classes. During the peak traffic period, traffic is scattered rather than concentrated in one location.

See Also:-Top Intelligent Video Technology Solution Companies

Read Also

follow on linkedin Copyright © 2021 www.appliedtechnologyreview.com All Rights Reserved | About us
Top