Ensuring Success

Make sure you have the right camera placement for optimal use of analytics

• By Ian Westmacott
• Aug 01, 2015

Those words are often associated with the real estate industry, but the security community can also apply that mantra to the all-important issue of camera placement for optimizing analytics.

One of the major knocks on analytics since its introduction has been that it doesn’t always deliver on expectations. End users want to gather all sorts of data ranging from license plate recognition to people counting to dwell and linger analysis, yet the results can be inconsistent.

The biggest culprit in creating patchy outcomes is improper camera selection and positioning. If this isn’t handled properly from the onset, the results from the analytics provided by these cameras will never meet expectations. No matter what else you do, if you mount a camera overhead and then decide you want it to read license plates, it can’t do the job. Similarly, a camera that is set up in the dead of winter near a parking lot surrounded by trees won’t be of much use when summer comes and the trees have grown and are now obscuring views, casting shadows and creating interfering movement.

Getting the most out of an analytics system has to begin long before the first images are captured. It starts with knowing what you want the analytics to do. This will then help determine the types and numbers of cameras used, the mounts on which they are installed, and the way they are positioned within various settings.

Fortunately, there are professionals within the manufacturer and integrator communities that can walk an end user through the process of selecting and installing cameras to maximize the analytics output. But, it is also important to know what the potential issues are for the different scenarios so you can make informed decisions.

There are many factors that impact video quality—illumination, the size of the asset, separating assets from people, obstructions, environmental issues and movement. We’ll take a close look at each of these and how proper camera positioning can work around them.

Illumination

Light is considered a good thing, right? However, too much light in the video frame, such as in the form of headlight glare or reflection from a body of water or glass, and suddenly the analytics within the camera can miss an event or misinterpret it as a false alarm.

To avoid this issue, the camera needs to be positioned so that it isn’t pointed at the source of the reflection, such as a store window or mirror. Polarizing filters can be added to the camera lens to deal with this, while thermal cameras can also be part of the solution.

Headlights, brake lights and any other type of moving light can cause blooming, which stops analytics from monitoring a situation. There are a couple of solutions to this problem, one of which is to use thermal cameras. Another is to position the camera so it captures video beside the vehicles, rather than in front of it.

While too much light can be a detriment to analytics, so too can insufficient light. Thus, it is important when positioning cameras in potential lowlight areas to consider using thermal or integrated day/night cameras as well as to explore ways to use artificial light to illuminate the area.

Shadows from trees, clouds or people can cause analytics to misinterpret activity, triggering alarms when none has occurred or, if the activity happens within the shaded area, it may not be visible enough for analytics to see it. Better lighting and thermal cameras are two options for improving this scenario as well.

Size of the Asset

Analytics likes to track bigger objects, so making assets a larger portion of the camera view will ensure that they are recognized properly.

With a fixed asset, such as a shelf full of merchandise, the best way to get the proper view is to move the camera closer or to adjust the zoom so the merchandise represents at least 5 percent of the camera view.

Separating Assets from People

While analytics can distinguish people from product, over time if the two occupy the same space within the camera view it becomes harder for the software to separate them. A person standing still in front of a shelf filled with merchandise, for example, could be mistaken as part of the display.

When he moves, it looks as if something is being stolen. Correct camera placement and proper use of analytics tools, such as drawing a detection region that includes the assets but excludes the people, can address this.

Obstructions

There are few scenarios that allow cameras to have unobstructed views from all angles, so it may be necessary to consider multiple cameras positioned for wide angle, overhead or other views to cover an area appropriately.

When placing the camera, you want to avoid blocking the view, either with fixtures, furniture, shrubbery or even people. At a cash register, rather than pointing a camera so it provides a head-on, blocking view, it’s better to position it with a sidelong shot that sees both the employee and the customer without obstruction.

Environment

Weather certainly plays a role in how analytics works. If there is constant movement from wind, rain or snow in a scene it can trigger false alarms. It’s best to avoid placement where moving branches are likely or where precipitation can directly affect the view.

Ultimately, we can’t control the weather and often we can’t control the environment either. In these cases the best we can do is to understand what impact the environment will have on the analytics program.

Moving Objects

Like trees swaying in the wind, all moving objects can cause issues. Therefore, when placing cameras, think about their proximity to anything that provides constant motion—escalators, automatic doors, even traffic that can be viewed through windows. If possible, aim the cameras so that uninteresting motion is not in the camera’s field of view. If this is not possible, then use detection regions to limit the analytics to areas away from the constant motion.

Having addressed the possible problems that can impact video quality, another factor to consider with camera placement is what you’re hoping to achieve with your analytics.

An activity such as people counting works best on a two-dimensional scale, so a camera mounted overhead works better than one positioned at eye level. With a straight on view, people coming into a building, such as those entering a stadium lobby might block others from the camera view. Positioned overhead, however, the camera can identify each person separately.

Dwell and linger is another great analytics tool that can be enhanced by proper camera placement. A retailer who wants to measure how many people look at a display and how long they stand there would need cameras that have an unobstructed view as well as one that provides enough space around the display to fully capture people as they move near it.

Think not only about what you want to achieve immediately, but also consider how needs may change down the road. If surveillance is the current plan, but at some point license plate recognition is a goal, consider sourcing and placing cameras with both needs in mind. While this is not always possible, a little planning up-front can often help down the road.

The science of analytics relies very much on the art of camera placement. But if the two are considered together from planning through execution the outcome will be a successful one.

This article originally appeared in the August 2015 issue of Security Today.