Join or Sign in

Register for your free asmag.com membership or if you are already a member,
sign in using your preferred method below.

To check your latest product inquiries, manage newsletter preference, update personal / company profile, or download member-exclusive reports, log in to your account now!
Login asmag.comMember Registration
https://www.asmag.com/project/micron_edge_storage_for_video_security/
INSIGHTS

With AI enhancement, recognition is more accurate, and applications are more diverse

With AI enhancement, recognition is more accurate, and applications are more diverse
With the support of AI, recognition becomes more precise, and applications more diverse. Thanks to recent advancements in artificial intelligence applications, the surveillance industry has greatly benefited.
Taiwan is the cradle of the surveillance industry. Over decades of evolution, the purpose of the surveillance industry has extended beyond "security" to include more advanced applications such as access control, retail management, parking lot operations, factory management, and medical applications, helping vertical industries create smoother and more efficient operating models.

However, as most surveillance operators have long focused on "surveillance" itself, applying it in vertical fields often requires piecing together solutions from different providers to meet the demands of property owners.

For example, automating license plate recognition for parking fee collection might involve a combination of cameras, peripheral actuator boxes, computer servers, or AI boxes. These devices often come from different manufacturers, making it easy to stack up various costs, such as integration and testing difficulties, troubleshooting uncertainties, and time-consuming processes, not to mention the potential disasters caused by the need for RMA repairs, leading to operational stops or other unforeseen calamities.

With the support of AI, recognition becomes more precise, and applications more diverse. Thanks to recent advancements in artificial intelligence applications, the surveillance industry has greatly benefited. AI computation's biggest contribution to surveillance is enabling more accurate image recognition and "behavior detection".

For instance, before the introduction of AI computation, surveillance systems could only use pixel differences to determine whether an object had intruded, remained, or disappeared from the scene, unable to accurately identify the object. A dog entering a restricted area and a plastic bag flying into the scene could easily lead to false alarms, wasting resources and alert processing efforts.

With AI computation, we can accurately identify object types, whether it's recognizing faces, car models, or license plates, nearly any object trained through deep learning can be recognized.

By installing AI directly in front-end cameras, a more streamlined surveillance framework can be achieved. Consider whether there's really space for additional wiring or server setup in places like road intersections, factory server rooms, or environments where production equipment operates.

Deploying an AI surveillance environment could be as simple as unplugging the old monitor from the network cable and replacing it with a new AI camera. Therefore, front-end AI cameras naturally became the mainstream in the AI surveillance market. Front-end AI cameras can perform the following tasks:
  • Direct AI image recognition on the camera, such as recognizing employee license plates in the factory area.
  • Direct event management and behavior detection on the camera, such as recording an event if an unauthorized license plate is detected leaving.
  • Directly control peripheral devices from the camera, such as opening the parking barrier for license plates on the whitelist.
Previously, complex wiring and server configurations were required to accomplish these tasks, but now a single AI camera can handle them.

Behavior detection is the real magic of AI image recognition. Merely identifying objects is not enough; it's also necessary to know "what is happening". For example, detecting a vehicle entering a factory area, can we tell if it's parked correctly or illegally on the grass? Regardless of the object being recognized, several behavior detection modes should be available:
  • Restricted area detection: Triggers an event when specified objects enter a restricted area (e.g., animals accidentally entering a track).
  • Time-limited restricted area: Triggers an event when specified objects stay in a restricted area for a set time (as above).
  • Line crossing detection: Triggers an event when specified objects cross a line (as above).
  • Lost object detection: Triggers an event when specified objects disappear from the scene for a set time (e.g., equipment taken from a lounge).
  • Lost/found object: Triggers an event when lost or found objects appear in the scene for a set time (e.g., unidentified items left in a lobby).
  • Tamper detection: Triggers an event when the camera's view is obstructed (e.g., by a balloon).
  • Simultaneous presence detection of the same species: Supports detecting more than one specified object in the recognition area, triggering an event if present at the same time (e.g., wearing safety helmets and vests simultaneously).
  • Absence detection of the same species: Supports detecting the absence of one or more specified objects, triggering an event if any are missing (e.g., not wearing safety helmets and vests simultaneously).
With behavior detection, AI recognition becomes meaningful, generating a rich combination of events for more versatile applications.

By not binding to models, the application scope of AI cameras is extended. Currently, AI models applied in industrial and commercial fields mainly focus on recognizing people, objects, and vehicles within specific areas, counting statistics, followed by license plate recognition for access control, and then controlling site safety with construction helmet and vest detection, smoke and fire detection, etc.

These models and functions can be directly provided by AI camera providers, while some AI cameras support models trained by users. For example, if a factory area needs to recognize specific objects or products, these models can be packaged into kits and loaded into AI cameras, eliminating the wait for official models from AI camera providers.

These AI functions can be easily applied in typical industrial and commercial factory scenes, such as:
  • Object recognition of people, vehicles, and objects: Designating specific areas, warning lines, or restricted zones in the factory area, recording or issuing alerts when specific personnel or vehicles enter.
  • People and vehicle counting: Counting the density of personnel and vehicles in offices, meeting rooms, or parking lots for crowd or traffic control. In retail spaces, this can serve as operational reference.
  • License plate recognition: Setting blacklists and whitelists for the factory area to control vehicle entry and exit. Trigger special event notifications for specific VIP license plates, such as welcoming.
  • Mask and safety helmet and vest detection: Detecting whether personnel in construction sites are wearing appropriate safety equipment correctly.
  • Smoke and fire detection: Not every environment in a factory area may be suitable for installing smoke detectors, so AI cameras' smoke and fire detection functions can prevent accidents before they happen.
  • Lost or found object detection: If an object is lost in the factory area, or if something that shouldn't be there appears, an alert is issued, helping management respond as quickly as possible.
Integrating back-end servers for video management and event retrieval becomes easier. After AI cameras identify objects and send out events, they can immediately interact with peripheral devices to respond to events, such as opening the parking barrier for whitelisted license plates. If the event doesn't require immediate processing, it can still be stored in the back-end video server for later retrieval.

The key is that using "events" as an index for search is much more precise than simply using time. For example, if a public facility in a factory area is found missing, without event records, it might not even be known when the item was lost, leading to a needle-in-a-haystack search through the possible timeline, undoubtedly wasting resources.

Video servers integrated with AI cameras can receive event records sent by cameras, including the time of the event. By categorizing video clips by event type, managers can quickly retrieve relevant video clips by accessing the corresponding event category.

In the scenario mentioned above, managers only need to access the "lost object detection" event records to quickly determine when and how the loss occurred.

Seamless integration with factory IT environments. For larger companies deploying surveillance environments in the past, one of the most painful aspects might have been integrating with the factory's IT environment. For example, how does the HR department's HRM system communicate with access control equipment? Information display in parking lots (such as available spaces and vehicle occupancy) often requires bypassing many systems to achieve. If the data exchange format of the front-end AI camera or the entire AI surveillance framework is closed, it naturally causes headaches for IT staff.

The ideal front-end or back-end surveillance system would provide open data formats, even offering SDKs or APIs for engineers to access and integrate data. IT staff can also follow this standard to develop back-end software for more customized applications, rather than being restricted by surveillance providers, which is a healthier industry model.

License plate recognition applications are a prime example. IT managers can access event data in cameras through SDKs and write their own rules to decide whether to display a welcome message for VIPs on the entrance LED screen. If surveillance providers do not offer open data formats, this task becomes very difficult and complex, undoubtedly adding costs.

Deploying AI surveillance environments with simple and trustworthy architectures. Previously, due to architectural limitations, introducing AI surveillance might have required overcoming significant barriers. As mentioned earlier, the simplest method might even involve simply removing the old camera from the network cable and plugging in a new AI camera, without affecting wiring or needing to deploy new lines.

Another important aspect is cybersecurity. If security vendors fail to ensure cybersecurity, it means going against the purpose and deploying even more dangerous systems. The IoT cybersecurity certification by the Taiwan Association for Information and Communication Standards (TAICS) can effectively safeguard this aspect, preventing data leaks, and should be a priority consideration when purchasing surveillance equipment.

Secutech Taipei

Secutech is the leading platform in Asia for connecting with regional partners in security, IoT, and AI. Register as a visitor today to discover the latest AI solutions for security, mobility, retail, smart home, factories and disaster prevention, all under one roof. Engage directly with manufacturers with the technical expertise to develop tailor-made AI-powered security and safety solutions for both public and private sectors. 

Event date: 24 -26 April 2024
Location: 1F, Taipei Nangang Exhibition Center, Hall 1
Website: https://secutech.tw.messefrankfurt.com/taipei/en.html


Product Adopted:
Others
Subscribe to Newsletter
Stay updated with the latest trends and technologies in physical security

Share to: