Guide to Dahua AI CCTV camera and recorder equipment

31 Aug

An introduction to Dahua AI CCTV camera and recorder (NVR / DVR) equipment, explaining how Dahua’s artificial intelligence technology works, and explaining the different AI functionality Dahua provide via their IVS technology.

AI CCTV Hardware – Deployed at the edge (On-site)

CCTV “AI enabled” hardware typically means that the hardware has a built in a AI computer chip that handles a set of algorithms (baked into the firmware of the product) designed to perform certain analytics tasks on video streams, such as Face detection or ANPR.
AI functions can be found in IP cameras and DVR/NVR recorders.
There are “general” AI CCTV cameras and “special” AI CCTV cameras. General AI CCTV cameras are designed to be deployed in a large number of scenarios. Where as “special” AI CCTV cameras are designed for specific scenarios such as Thermal Body Temp cameras or Face detection cameras and are supposed to give the best results for those scenarios.
AI functions inside “AI branded” IP cameras are useful to either push an alert message to another device (I.e Smartphone/VMS software) or control something like a door/alarm upon a certain event occurring without the need of full CCTV network. The latency of result is in the nano seconds, as all the processing of data is done on the camera before been transmitted. However, “good quality” AI camera hardware is quite expensive, especially when the application requires a high accuracy rate/special features such as body temp monitoring.
AI enabled DVRs and NVRs are useful because they provide a way of applying multiple different AI functions to create alerts on connected video feeds for different scenarios. They as well as give the ability of viewing and storing for captured data via monitor, as well as having a more complex set of algorithms and better more powerful hardware. Typically NVR recorders are used in combination with AI IP cameras to provide a secondary compute layer, giving the highest accuracy rate achievable on a given function (As long as both camera and NVR supports the same function). As the event is checked by both systems simultaneously. Most vendors/high-end manufactures recommend AI camera + AI NVR setups.
COAX AI camera’s are not available, as traditional cameras only provide AV streams. AI features are only available on COAX camera streams when connected to a DVR.
There are hundreds of AI CCTV cameras and recorders available from Dahua.
All products (be it cameras or recorders) differ in what AI functions they support and what scenarios they are useful for.
The performance accuracy at a given task also differs between products. The more expensive units typically perform with higher levels of accuracy due to better/more powerful hardware components.
Some Dahua AI functions include the ability to: detect faces, people count, ANPR, object tracking and monitor people’s temperature.
Dahua’s term for AI “features” is IVS (Intelligent Video System).
IVS functions are marketed and broken down into 3 classes. They are: Premium IVS, Advanced IVS and Analytics+.
Premium IVS functions are the “basic” AI functions available on most of Dahua’s AI camera/recorders range.
Advanced IVS & Analytics + functions are hardware specific. As these functions require more processing power and data from a video stream/scene.

See below for a full list of IVS function and definitions:

Tripwire – Premium IVS
Tripwire can only be applied to the scene where the targets are sparse and there is rarely any covering between targets, such as the perimeter without any guard.
Function Description: it can be used to detect if there is any object crossing the warning line, it can support different direction selection, such as A->B, B->A and bidirectional, and it can activate record , snapshot and alarm according to the judgment result.
Intrusion - Premium IVS
Intrusion Detection is to detect if there is any object entering or leaving the area, it will trigger rule when crossing the area perimeter, support enter, leave and enter&leave three direction selection, it can activate record, snapshot and alarm according to the judgment result.
Application Scene: It is similar to warning line. It needs to leave some space for target movement outside the perimeter if it is to detect enter and leave event; It can only be applied to the scene where the target is sparse and there is rarely any covering between the targets, such as the perimeter without any guard. It will trigger alarm when the target is entering or leaving the area.
Abandoned &Missing object detection - Premium IVS
Object abandoned means that it is to detect if there is any object abandoned in the designated detection area and surpass the set time. It will activate record, snapshot and alarm according to the judgment result. Object missing means that it is to detect if there is any object moved in the designated area and surpass the set time. It will activate record, snapshot and alarm according to the judgment result.
Application Scene: It can be applied to the scene where the target is sparse and there is no obvious or frequent light change. Missing alarm will increase in the scene where target density is high and there is frequent covering; false alarm will increase in the scene where there are too many people. As for the detection area, it needs to be simple; it is not suitable for the too complicated area.
Scene Change - Premium IVS
Video scene change function is used to detect if the camera is covered or moved and other abnormalities of monitoring scene change, such as a camera being blocked, moved, intense light change, etc. The intelligence base (algorithm) will compare the images before and after being covered or turned to judge if there is any similarity; if not it will generate an alarm.
People Counting - Advanced IVS
It is to make statistics of enter and leave people in the specific area of the image, and finally it comes out the statistics data, providing kinds of data reports such as day report, month report and year report. There are two ways to present: line chart and bar chart, providing data exporting function.
Loitering Detection - Advanced IVS
Monitors a defined area for the presence of an object that remains in the area longer than a specified amount of time.
Fast Moving - Advanced IVS
Triggers an alarm when the speed of an object exiting a defined area exceeds a specified sensitivity setting.
Crowd Gathering - Advanced IVS
Monitors a defined area for a set number of people gathering and remaining inside the defined area.
Parking Detection - Advanced IVS
Monitors a defined area for the presence of a vehicle that remains in the area without motion.
Heat map - Advanced IVS
Heat mapping is an advanced feature that brings business intelligence to a security system. Intelligent analysis tracks and stores daily, weekly, monthly, and yearly people flow through a defined scene and produces a configurable report.
Uses collected data to produce a heat map, a two-dimensional representation of data that provides an immediate visual summary of information Identify the most active areas of a given location, providing valuable data analysis for commercial applications.
ANPR - Analytics +
Body temperature monitoring – Analytics +
SMD - Premium IVS
Smart Motion Detection (SMD) uses an advanced algorithm to differentiate human and vehicular shapes within a scene and send alarms only when a person or vehicle is detected. It is applied by one click process and doesn’t require any rules to be manually set.
Face Detection & Recognition - Analytics +
Uses deep learning algorithms to create positive IDs individuals. Once face detection technology identifies a human face and captures the image, the image is transformed into a vector object (also called face signature), based on image landmarks.

A list of Dahua AI enabled cameras and recorders is available by clicking here.

AI analytics services in the cloud

What is AI VAS (Video Analytics services) in the cloud?

AI “cloud” VAS is a way of achieving the same or similar functions as above (such as Face detection) while outputting data with much higher levels of accuracy, as they leverage much more advanced/powerful hardware and a much more complex set of algorithms than any recorder/cameras working at the edge.

One way to run Cloud AI VAS is through a SaaS model from Amazon, IBM and Google that provides most of Dahua’s IVS functions (such as Face recognition) and more using any 3rd party ONVIF camera/recorder but is very costly to run in real time. Therefore, snapshots are best used and uploaded to the service in regular timed intervals instead, which limits the applications it can be used for.

SaaS models can be costly due to them being charged per API call, which equates to every time a function is used – such as if a person walks in from of a video feed and it was connected to Amazon Recognition and programmed to detect people. That person would trigger the “detect person” function and would cost a chargeable API call like a telephone call to the service, requesting that the service looks at the stream and ID if it’s a person or not. Even if it’s an error – it still costs.

Another way of running cloud AI VAS solutions is by writing our own server-less application leveraging freely available open-source algorithms, “world class” pre-trained datasets/models and frameworks to run in the cloud.

There are many advantages of using a Cloud CCTV solution such as:

Eliminate the NVR/DVR as single point of failure and security risk
Lower hardware support costs
Push updates and security patches automatically to each camera
Alert on tampering or camera malfunction
Use the power of Cloud based analytics such as: index and activity based search, ANPR, facial recognition
Easy sharing of cameras across all platforms and mobile devices and between agencies
Simplifies network and security configuration with no need for complex VPNs
Fully encrypted data at rest and in transit
Centralised remote-access on command that is scalable to any number of locations.