The Hidden Costs Behind Motion Detection

Computationally speaking, motion detection is a relatively cheap process that can even be performed by low end devices. This is deceptive because the ineffectiveness of Motion Detection actually incurs many other costs down the line.

In our experience of applying AI to analyse literally billions of images, we’ve discovered that conservatively only 10% of all security camera footage contains info that is interesting to the end user. In other words, 90% of camera footage that is stored in the cloud is noise that can be safely discarded.

This leads to various cost factors:

Cloud Storage costs

The cost of storing data is the most apparent cost when maintaining a cloud camera platform, since you’re dealing images and videos. While the cost of cloud storage is getting cheaper and cheaper, the amount of data generated by security cameras far outweighs this benefit.

Doing some back of a napkin calculations:

In a scenario where 10,000 cameras are uploading only 1 hour of footage each day for a month, and assuming these cameras are uploading 720p video at a bit rate of 5Mbps.

5Mbps * 1 hour a day * 30 days * 10,000 cameras = 675 terabytes

Storing 675 terabytes on AWS S3 would cost approximately $16,000 a month.

Using Image Intelligence as a noise filter would reduce this to approximately 10% of the initial size.

Storing 67.5 terabytes on AWS S3 would instead cost $1,700. An order of magnitude less.

These numbers are also conservative:

  • The cameras would probably record more than 1 hour day, since motion detection is broken.
  • Some cameras stream footage 24/7 to the cloud.
  • 10,000 cameras are a small number of cameras
  • The bit rate could be even higher if recording higher quality video

Data transit costs

Not as apparent but related to storage costs are data transit costs. Generally any inbound data transfer is free, however any outbound transfer incurs a cost. In other words, receiving all the footage from thousands of cloud connected cameras is free of charge, but as end users view this footage, data that then flows back out to the end users will incur fees with your cloud provider.

Continuing from the above scenario, let’s imagine that only half of the footage is viewed by users.

675 terabytes of total footage / 2 (half of content actually viewed) = 337 terabytes

337 terabytes data transfer out from AWS S3 = ~$20,000

Now if Image Intelligence was used as a filter

67.5 terabytes of total footage / 2 = 33 terabytes

33 terabytes data transfer out from AWS S3 = $3,000 (85% saving)

User experience cost

Least apparent of all would be the user experience cost. Motion detection ultimately produces too many false positives. The end user is thus bombarded with too many notifications, leaving her with two undesirable outcomes:

Respond to each notification by viewing the footage.

This is a huge distraction and waste of time. The user would most likely then adopt the next strategy.

Ignore the notifications and risk missing important events. If there is something important that has happened, review all the footage around the time of incident.

The time it takes to find a needle in the haystack is not only costly but also stressful if a horrible incident has occurred. Being delayed in responding to an incident could also result in irrecoverable loss.