Solving Video's Searchability Problem
Video is the fastest growing type of data in the world. According to Cisco's recent Visual Networking Index Report, “It would take an individual more than 5 million years to watch the amount of video that will cross global IP networks each month in 2019. Every second, nearly a million minutes of video content will cross the network by 2019.”
Additional research by IDC has shown similar findings. The firm reported that in recent years, surveillance video has become more than half of global big data, and this percentage is forecasted to grow to 65 percent by the end of 2015.
The amount of information contained in all this video is nearly endless. Who is in the video? What are they doing? Where are they? Does anyone even know? Unless the person responsible for the upload has taken the time to write a highly detailed description of the entire clip, much of the content — and context — ends up in an unsearchable black hole. That is also assuming this person, or team of people, have been able to describe all the live or recorded video.
Let’s say a large enterprise with 100,000 cameras worldwide records just one hour of video per day (as opposed to the 24 hours they are likely capturing) — that is just shy of 37 million hours of video per year. There is no way a team of people can review this amount of content.
How can your enterprise customers handle this kind of volume? Software is the answer.
Even though video represents the largest and fastest growing data type, companies have yet to figure out how to index and search the massive volume of content in a meaningful way. No matter if the video is originating from a live feed or has been previously recorded, companies are relying on manually entered descriptions or basic thumbnail reviews to locate videos of interest. In a time-sensitive investigation or emergency situation, it is impractical to expect human reviewers to sift through all the content and effectively identify all the events of interest, if any.
When it comes to security and surveillance applications, many VMS platforms attempt to solve video’s discoverability problem through basic thumbnail and timeline searches; however, this is only part of the solution. Timeline searches still require users to know approximately when an incident occurred. The greater the possible time range, the more video there is to sort through in order to identify suspects or other important information. This can be frustrating — not to mention it costs valuable time.
Because humans and basic timelines searches have inherent limitations, software is the answer to significantly augment this process. It is much more useful for video to be instantly searchable based on the temporal and spatial events contained in each scene and derived from automated analytics. Then investigators can re-direct their time evaluating the evidence found as well as the broader case management activities.
Fortunately, today’s most advanced video management systems are finding ways to look withinsurveillance videos and make the visual content searchable based on any number of specific criteria, including the basic thumbnail review. For example, users can pull up all video clips related to a red car, traveling west, with a specific license plate number. Or, they could instantaneously find videos of a child, with black hair, wearing a pink jacket and exiting the building.
The ability to supplement the manual act of scrolling through video — and directly locate precise areas of interest within surveillance video content — is made possible by analytics such as facial recognition, license plate recognition, dwell time, motion and others.
While some VMS present search and analytics as two separate user experiences, the more advanced VMS providers are now combining these into one intuitive user-interface. Essentially, data from any video analytic becomes the metadata foundation used to search for and precisely locate relevant video clips. In addition, the search becomes even more sophisticated when the analytics are combined with data from systems such as access control, point of sale (POS), ATM and other third-party systems that have been integrated into the VMS.
For example, the power of combining analytics with existing systems can be demonstrated using the previous example of the red car and the child with black hair. Video analytics are used to identify various shades of red and pink, discern a child from an adult and read license plate numbers. Then, data from the access control and POS systems can determine who entered a location and who made a purchase. VMS search engines can then synthesize this data into useful search information to help your customer quickly and accurately locate relevant videos and events. The process is reliable and efficient, and much faster than traditional search methods.
Better Alerts and Case Management
In addition to forensic search, a VMS that uses analytics to improve search results can also deliver proactive alerts that are more reliable than those from the typical VMS. One 3VR client has been using advanced search and analytics to keep forklifts out of restricted areas. Occasionally a forklift driver will cross over a boundary, which creates a safety hazard for other employees. Many VMS platforms can only alert security personnel based on a trip line or a motion incident within a given area; however, in this case, the customer found that this was creating too many false alarms, particularly as other employees and machinery were not restricted in the same manner as the forklifts. By using advanced search criteria, the customer can be alerted to the presence of a “large, moving yellow object” within the restricted area, which has dramatically cut down false alarms.
VMS search based on video analytics also enables better case management. It speeds up the ability to find, save, share and review cases — whether they are isolated incidents or ongoing investigations. Cases can also be shared or private, enabling multiple investigators to work the same case from different locations, or limiting sensitive information to just one or two qualified managers.
Video Discovery vs. Video Viewing
Using analytics as the cornerstone of search within a VMS is currently the most effective way to solve surveillance video’s “searchability” and “discoverability” problems. It is game-changing technology for the security industry, and it is revolutionizing how dealers, integrators and end-users think about and utilize video.
It is bringing structure to previously unstructured video content in a way that text-based searches or timeline searches never could, making the growing volume of video data much more accessible. Ultimately, it is turning the VMS into a true data management system that makes video discoverable, rather than the “video viewing platform” that previous VMS versions tended to be.