According to recent industry reports, the volume of daily data generated by surveillance cameras worldwide exceeds 850 petabytes in 2017. By 2019, this is expected to grow beyond 2500 petabytes. This is equivalent to 50 million double-layer Blu-ray movie discs.
Further, consider this.
By 2020, over 200 million IP cameras are expected to be shipped annually. A single surveillance IP camera will generate ~60 GB of HD video data per day. This amounts to storage space requirement of >5.3 TB for 90 days (typical archival period for surveillance systems). Cost wise, this turns out to be ~$135 (hard disk storage) or ~$160 (cloud storages such as Dropbox or Google drive) per camera. The overall costs goes further up when one considers transmission bandwidth (to upload to cloud storage or servers for analysis). What’s more, there is an enormous expense of human resources needed for visual inspection of the entire video footage to look for incidents of interest.
That sure is a lot! So, what are some ways to bring about a reduction in Operational Expenditure (OPEX)?
As you may have read in our earlier blog, savings can be achieved by using ‘intelligent’ DVR/NVR with built-in visual analytics at the edge.
But how does one bring in the said ‘intelligence’ to a DVR/NVR edge device?
Well, the solution is 3-pronged. Let us try to understand this in more detail.
(ONE) Encode with Intelligence: Typical surveillance videos have long periods where there is very little or no activity. During such periods the DVR/NVR edge device can considerably lower the encoding quality (by lowering the resolution, frame rate and bit rate for encoding) to achieve huge savings on storage space and/or streaming bandwidth. This is known as context-aware encoding. For instance, the DVR/NVR can build in intelligence to identify moving objects of interest (ex: people, cars) using computer vision or deep learning and switch back to high quality encoding when a positive identification is made. The plot below illustrates bit rate savings that can be achieved with intelligent encoding. Savings in the range of 50–70% can be expected for video surveillance.
With such a scheme, the key to maintain reliability of surveillance is to identify objects of interest very quickly to ensure that the encoding quality involving their presence are recorded/streamed at the desired video quality. This is vital for using these recordings for evidence management & law enforcement.
(TWO) Index with Intelligence: The eventual goal of recording video is to later retrieve it for finding events/people/objects of interest. This is often time consuming. The intelligence here comes in the form of automated means to analyze and index DVR/NVR recordings with metadata/tags pertaining to such elements in video. For instance, video can be analyzed for presence of cars and bags, and for events such as loitering, camera tampering or intrusion detection. The metadata/tags can be stored along with other information such as date, time, GPS coordinates etc. to lend itself well for law enforcement. Deep learning is often used for such tagging in background, and can happen even while other recordings are in progress so that tags are available soon after the recording is available on the DVR/NVR storage.
(THREE) Retrieve with Intelligence: The amount of time required for visual inspection of the entire video footage can be drastically reduced if the user can easily search the footage for objects or particular events of interest, or use a combination of rules that will otherwise be virtually impossible for a human to search based on. Leveraging the metadata created via intelligent indexing, DVR/NVR edge devices can support intelligent search and retrieval options, along with smart visualizations (ex: highlighted objects and events, heat maps, object trajectories) to help users in their inspection. The key here is to provide flexible and versatile search options that provide quick and accurate results to the user within seconds even while combing through terabytes of data.
Leveraging all these three modes of intelligence, Ittiam has realized an Intelligent DVR/NVR solution – limeRecordPro. limeRecordPro brings in up to 70% savings while recording 4K UHD H.265 (HEVC) and H.264 video, supports low delay (<10 sec) tagging of objects of interests and compliments these with very quick search which makes results available within a fraction of a second.
limeRecordPro can be used for edge devices in surveillance, medical, enterprise, industrial and defense applications. Also check out adroitVista SDK – Ittiam’s suite of modules for realizing high-performance machine vision, learning and analytics functions on edge devices.