Assessing Video Search (Part 1)
Video has become an increasingly proactive tool, with many users opting for solutions that use triggers and incidents detected by video analytics and other smart algorithms to initiate actions and system events. Despite this, recording video remains a vital requirement for the vast majority of video users. While integrators and end users put a lot of effort into specifying video management systems that capture high quality video, too often less emphasis is placed on retrieving video of interest, quickly and easily.
Smart options for video surveillance are being developed at a rapid pace. As systems have increased capacity to use captured metadata to initiate actions and alarms, to trigger peripheral systems and to collate and report on business intelligence, many of the benefits on offer from modern technology do not require operators sitting in a control centre, watching screens for signs of exceptional behaviours.
There can be no doubt that modern video surveillance technology is a great enabler, and the growth of AI deployments have only served to enhance the possibilities on offer. Modern algorithms can exploit GPU architecture to automate a wide range of processes and tasks which previously had to be manually controlled.
In the past, it was not unusual for an operator to view video streams of vehicles arriving at access control points. After visually verifying the type of vehicle, and often either communicating with the driver or checking a list of expected vehicles, they would grant access by opening a barrier or gate.
Today’s smarter systems can simplify such tasks. ANPR can identify company or registered vehicles and allow access. Sometimes the process will be straightforward, simply using a rule to open a gate. On occasions, a video clip might be bookmarked and linked to the vehicle registration number. There is no need for operator intervention.
Equally, using object recognition with deep learning-based systems, the analytics engine will recognise certain types of vehicles, using not only ANPR but also the image content, to ascertain if it represents a risk or not. Where the system is satisfied there is no threat, access can be granted. If there is an issue, but it is not security related, a depot manager – for example – could be notified. Again, there is no need for operator intervention.
There are hundreds of examples where video surveillance can be automated to carry out more menial tasks, only involving the security team if a threat is detected. Whilst these features add value for the end user by delivering everyday benefits, the reality remains that users expect their surveillance systems to archive footage so if an unexpected incident occurs, or verification of an event is needed, further investigation can be carried out.
Even the smartest video surveillance systems out there will include video recording. However, where end users have invested in an advanced video solution capable of delivering smart functionality, they rightly expect searching, finding and accessing relevant and critical video to be simple.
When an unexpected event or incident occurs which has not been legislated for, often the only way to mitigate risk and address threats is through an operator investigating footage and taking appropriate action. If this is the point where time is lost and a response delayed, then many end users will consider their solution to have failed, and subsequently their investment in a smart security solution to be wasted.
While there are a growing number of intelligent implementations that can be created with modern video surveillance technology, the reality is that unless the simple things, the fundamental basics, of video surveillance, are efficient and effective, the end users’ expectations will not be met. Searching for, and finding, critical video quickly is an essential function. If that does not work well, then other intelligent features won’t make up for the poor performance.
ACC (Avigilon Control Center) is an open platform VMS available in three variants: Core, Standard and Enterprise. The search capabilities differ dependent upon which licence option is chosen, with Appearance Search included in the Enterprise edition of the software.
The main differences between the three editions in terms of device support are the maximum number of cameras: 24, 75 and in excess of 10,000 respectively. The versions allow 24, 75 or 300 cameras per server, and both Core and Standard are single server implementations, with Enterprise supporting over 100 servers in a system.
When it comes to search capabilities, the Core version – which is the entry level option – has a decent level of search criteria and reflects what you’d expect from a mainstream VMS. The options are motion detection filtering, intelligent motion search, event search and thumbnail search. When compared with NVRs offering an equivalent level of performance, these options do allow a higher degree of flexibility.
Whilst fairly standard in terms of functionality, the search options are well implemented and deliver fast and accurate results. The timeline is intuitive, making operations simple.
With the Standard license, the search options are enhanced to include searches based upon data from ACM, Avigilon’s access control manager, and data from a linked POS terminal. These additional searches will be of interest for some users.
The Enterprise level software is the choice for users who need to ensure quick and accurate searches across large systems. This includes all the functionality of the Standard version but also allows Alarm and ANPR searches, along with the powerful Appearance Search technology.
Appearance Search uses AI and deep learning techniques to provide smart searching. It speeds up the process by accurately filtering hours of video, identifying matches across multiple streams in a few seconds.
To access the power of Appearance Search, integrators and installers will need to run ACC Enterprise (version 6 or later) on an Avigilon NVR. Legacy NVR 3 and NVR 2 models require a GPU upgrade kit, but the NVR 4 series makes use of the latest processors.
It is also necessary to deploy Avigilon cameras with self-learning analytics, or if legacy third party cameras are used the AI Appliance is required.
Appearance Search doesn’t require any configuration by the system integrator or security installer during installation.
ACC has an integral object classifier algorithm, and this automatically detects people and vehicles, applying a bounding box. These boxes are key to the analytics which power Appearance Search. Instigating a search is simple and can be activated with minimal data input.
When viewing a live or recorded stream, clicking into the bounding box brings up a contextual menu with options to initiate a search either before or after the point in time of the viewed footage.
For example, if an object has been removed from a storeroom, the user identifies points where the item is present and then missing. The search will deliver a sequence of thumbnails, allowing the moment the item was removed to be identified. From the relevant clip, the operator can click into the bounding box of the person removing the item, and by selecting the appropriate search, other footage of that person can be found, even across multiple cameras.
If searching for a person, the user can enter information about the target such as gender, hair colour, clothing colour (upper and lower body colours can be specified, and multiple colours can be selected), approximate time and date and the cameras to be searched. If a location is known, specific cameras can be selected, or a search across all cameras can be carried out.
Initiating the search is very simple, and filtering the results to achieve high levels of accuracy is equally quick and easy.
When an Appearance Search is run, the system presents a series of thumbnails based upon the search criteria. If the thumbnail contains a person in the background, the presented image zooms in on the facial details to assist the operator. From the suggested thumbnails the operator can view the linked footage with a single click, and if the target is verified as being correct, the thumbnail can be ‘starred’ via a single mouse click. This action informs the search algorithm that the target is valid. As more thumbnails are starred, the algorithm has more instances to refine the search criteria, and subsequently the accuracy of the search increases.
As thumbnails are starred, an icon appears in the timeline at the bottom of the operator screen, identifying the various clips. This enables an ‘at a glance’ understanding of the chronology of an event. If the clips are replayed, the system will show them in the correct chronological order.
A bar graph appears as the search is carried out. The length of the bars indicates the likelihood of a valid target being included in that specific time window.
Searches are fast, and as thumbnails are starred the GUI dynamically presents thumbnails with a higher likelihood of containing the target.
Managing search results is straightforward. Numerous starred clips can be selected and archived to create an archive of the event. The archived footage includes a standalone player which generates an ACC interface. This ensures that those without access to ACC can use the archived searches.
Appearance Search is fast and accurate, and is an ideal tool for busy applications with large camera counts. It currently sets the bar high for other VMS packages in terms of searching, and whilst the increased use of GPUs is seeing a lot of development in terms of search options, Avigilon’s ACC delivers the functionality today, in a well delivered implementation.
Milestone: XProtect Corporate
Milestone XProtect Corporate is the company’s flagship VMS. It is best to consider it as a unlimited solution. It supports an unlimited number of cameras per server (obviously server capabilities will ultimately dictate how many devices can be supported without an impact on overall performance), an unlimited number of severs, and an unlimited number of users. The video client for system control is free, and so offers freedom with regard to system management. Mobile connectivity is also supported.
XProtect Corporate supports a wide range of video formats including H.265, H.264, MPEG-4, M-JPEG and MxPEG. The solution is compatible with devices from over 100 manufacturers, including the vast majority of the manufacturers of professional surveillance equipment. The XProtect software has also been used for integrations with other systems to offer bespoke solutions, and so offers enhanced interoperability.
Features include support for failover servers, scheduled archiving, interactive mapping, hardware acceleration, support and management of edge storage devices, as well as integration with system elements including analytics, ANPR and video wall management.
Video search options are accessed via the Milestone Smart Client. Searches can be carried out based upon time and date, bookmarks, motion, alarms, etc., and are accessed via the Playback tab. There is also a Smart Search option which is accessed via the Live viewing tab.
While XProtect has a depth of functionality, it is fair to say that the search options are fairly standard.
That said, the company has announced it will be upgrading the search functionality. This will include the use of AI and deep learning technology. However, initial indications seem to be that this capability may not be accessed from the Smart Client as a standard feature, but will be dependent upon developer partners creating plug-ins to deliver enhanced search options.
The new search options will focus on making finding and tracking targets simpler and more intuitive, allowing data to gathered and used to increase the accuracy of searches. However, currently this functionality is still in development.
Currently, standard searches are initiated via the playback screen. Time and Date searches make use of the timeline. This is flexible and allows the timeline display to be adjusted to show varying time windows, ranging from four weeks to five minutes. Start and end times can be set, and a time window can be created by dragging and dropping. With recordings, alarm events and bookmarks highlighted in the timeline, this simplifies searches.
Another alternative is to make use of the Recording Search window. The timeline is used to insert a start time, and a search can be carried out across all cameras or selected cameras only. The search then displays displays the date and time of the first image in a sequence, marked by a green flag, the last image marked by a chequered flag, and the motion detection event which triggered the recording, marked with a yellow flag.
Highlighting a sequence displays it in a pop-up window.
More advanced searches are carried out via the Sequence Explorer tab. This gives two options: Sequence Search and Smart Search. The latter can be accessed from the Live tab as well.
Sequence Search firstly needs a time window specified, and then the cameras to be searched are selected.
Up to 100 cameras can be searched simultaneously. A frequency for thumbnails can be set: for example, this might be every 30 seconds or every 5 minutes, and is fully customisable. This delivers a thumbnail image at every specified time point.
With Smart Search, the camera is selected, albeit with the image masked by default. By selecting an area of interest, this unmasks that portion of the image; it is possible to add masking if the selected area includes a source of motion that is not of interest. A time window is then set, along with a motion threshold, and the search is initiated.
Smart search works well, and is simple to configure. While many VMS package include motion-based smart searches, the Milestone implementation is intuitive and easy to use.
The combination of searches cover most needs, and in the past XProtect has been one of the more user-friendly VMS options. Searching is simple and relatively fuss-free, and results are accurate.
There is a ‘but’; ACC has taken searches to a new level, and in doing so has addressed what was a serious issue for many end users. While Milestone is developing new search options, today’s VMS offering can’t compete for users who must have the ability to quickly find suspects across multiple cameras.