17,000 Cameras in Korea Wasted?
A Korean newspaper bemoans the waste of money spent on 17,000 surveillance cameras. The cause? "97 percent [of cameras] are unable to distinguish the faces of people or license plates of cars entering the grounds." While this may sound bad, the rationale used and specifics claimed misunderstands key surveillance technologies.
Let's first review the claims / asssumptions the newspaper makes:
- "[CCTV in schools] project started in 2009, and now there are some 100,000 of them in 11,000 schools, or an average of nine for each school."
- "It takes a camera with a resolution of more than 500,000 pixels" to distinguish faces or license plates
- "A security camera with a 400,000-pixel resolution costs W200,000 [~$200 USD], and one with 500,000 costs around W300,000 [~$300 USD], and the price tag rises to W2-3 million [$2,000 -$3,000 USD] for over 1 million pixels. But that is a price worth paying in schools where there is a lot of coming and going." [Note: 1 US = ~1000 Won]
- "30 percent of the 1,707 schools the BAI checked have CCTV cameras that face the wrong way or are blocked by trees or other obstacles"
- "Two out of every 10 schools put the monitors in the night watchman's room so they could not be viewed during the day."
- "Analysis of the various measures that have been taken to make schools safer, including security officers, police patrols and telephone hotlines, has found that CCTV cameras are the most effective."
There are some obvious and fair criticisms made - especially with cameras being poorly positioned and recorders not properly set up for 24/7 viewing. Basic design steps can and should easily correct that.
The megapixel claims are far trickier. While more pixels help, there is no magic number that ensures faces or license plates will be distinguishable. It depends on the FoV and the lighting. Even with 1 million pixels, the newspaper's recommended number, even in a best case scenario, 25 feet wide would be the most one could expect (assuming an aggressive 50ppf). This still would not be enough for parking lots and larger indoor areas.
Additionally, the pricing differentials are hard to justify - with the assumption that a 1MP cameras is 10x or more that of an SD (i.e., 400,000 pixel) one. Today, the prices are far closer with 2x being a better rule of thumb (see Korea's biggest manufacturer's recent MP offerings). As such, justification for new MP cameras should be far easier than they imply.
However, the expectation that all, or even most, cameras can capture facial details and license plate numbers is troublesome. Even with higher resolution cameras that expand the range and percentage of objects identified, we are still years away from being able to capture such details in nearly every spot of larger facilities like schools.