Cameras you can count on
Artificial intelligence (AI) and deep learning for video has opened the door to new possibilities where computers can analyze both live and recorded video to produce surprisingly accurate data.
By Colin Bodbyl
Information such as colour, quantity, frequency, gender, age, and more can all be collected automatically through AI and reported back to users. Previously, collecting such information would require a person to manually review hours of footage and document each data point.
There are two primary methods developers are using to present and utilize this data. The first is to present the data in a table as a report that outlines information about the people or objects the system observed over a set period of time. Retailers, for example, can use this data to see customer demographics in their store. They can analyze the average age group of their customers, their gender, or the colour clothing they wore.
While this data may be fascinating, it is challenging for users to generate revenue or reduce costs simply by having the data. Retailers still need to take action based on the data. This could mean changing their store layout or adjusting the products they sell. In addition, retailers then need to analyze how these changes affected their sales and finally they can connect that return on investment back to the data they originally collected from their surveillance system.
It is a long and challenging path for users to take action on this data and prove a return on investment in AI technology. How the data is used is key to proving its value, but without users taking further action the data is essentially useless. This requirement for further action, which users need to solve on their own accord, is a significant barrier to user adoption.
The second application of AI is to use the data as a tool to help users locate specific objects in recorded video using keywords. Most commonly we see this type of technology marketed as a tool for speeding up the search and video review process. Using AI, users are able to search video footage in a similar manner to how you would search the Internet using Google. A user could search for a white pickup truck travelling westbound and the AI could process the recorded video and show only events that show a white pickup truck travelling west-bound. At first glance this may seem extremely useful to many users.
The unfortunate reality is that in the majority of cases where users are searching through video to find an incident, they do not know what the suspect looks like, or the vehicle, or any details on the event for that matter. In addition, the current process of searching through video by simply skimming through motion events is already very efficient.
In rare cases, law enforcement may find themselves searching through hundreds of cameras for a specific object where this would be helpful, but the applications are few and far between.
Artificial intelligence will undoubtedly change our industry in the upcoming years. For this to happen, however, manufacturers and developers will need to find ways to make the data actionable and generate a tangible return on investment for end users. Current solutions are expensive and without a simple method to prove the technology is profitable, users will not adopt it.
Colin Bodbyl is the chief technology officer for UCIT Online (www.ucitonline.com).
This article originally appeared in the March 2018 issue of SP&T News.