Video analytics turn a camera into a detector: software watches the image for people or vehicles entering a defined zone or crossing a line, and raises an alarm with the video attached. Because the same camera detects and shows what it detected, analytics can make a perimeter faster to assess. They work well when the view, lighting and background are designed for them, and poorly when they are an afterthought.
How perimeter analytics work
Analytics software processes each video frame to find objects that move against the background, then classifies them, typically as a person, a vehicle or something else. Current products commonly use machine learning models trained on large image sets, which are much better than older motion-based methods at ignoring shadows, rain and swaying branches. The software then applies rules that the integrator sets up for each camera:
- Zone intrusion: a person or vehicle enters a defined area, such as the strip of ground inside a fence.
- Line crossing: an object crosses a virtual line, optionally only in one direction, such as from outside to inside.
- Loitering: a person stays in an area longer than a set time.
- Schedules and conditions: rules active only after hours, or only when a gate is closed.
When a rule is met, the analytics send an event to the video management system or alarm monitoring platform, usually with a short clip or a snapshot showing the object outlined. An operator verifies it and decides on a response. The analytics may run in the camera itself, on a server or appliance, or in a hosted service.
Visible and thermal cameras
Analytics can run on ordinary visible-light cameras or on thermal cameras, which form an image from heat rather than light. For perimeter detection the two have different strengths:
| Visible camera | Thermal camera | |
|---|---|---|
| Needs lighting at night | Yes, or infrared illumination | No |
| Detection range | Shorter for the same cost | Often longer, with strong contrast between people and background |
| Detail for identification | Good, with enough pixels and light | Poor; shows shapes, not faces or plates |
| Affected by | Darkness, glare, headlights, shadows | Heavy rain, fog, and scenes where background and people are at similar temperatures |
A common arrangement uses thermal cameras along the perimeter for detection and visible cameras, with suitable lighting, for assessment and identification at gates and approaches. Lighting design is covered in lighting for video surveillance.
Designing the views
Each camera has a range where it can reliably detect a person and a shorter range where an operator can see enough detail to assess what is happening. The detection zone drawn in the analytics should stay within the reliable detection range, and the areas that cause false alarms, such as roads, trees and neighbouring property, should be kept out of the zone or masked.
Other points that decide performance:
- Pixels on target. The analytics need a minimum number of pixels on a person at the far end of the zone, set by the product. Our guide to camera placement and pixel density explains how to work it out.
- Stable mounting. Poles that sway in the wind make the whole scene move. Use rigid poles or walls, and check them in a strong wind.
- Overlapping views. The ground right under and in front of a camera is often a blind spot, so the next camera along should cover it.
- Glare and headlights. Avoid views into the rising or setting sun and into vehicle headlights where possible.
- Infrared illuminators. Insects and spider webs near a camera's built-in infrared light show up as large bright objects at night. Separate illuminators, or thermal cameras, reduce this.
- Snow and seasons. Snow cover, drifting snow and leaves change the background. Review and adjust rules seasonally.
Pairing analytics with other sensors
Analytics pair well with fence and buried sensors, because their nuisance causes are different. A fence sensor alarm can call up a camera for assessment, while analytics catch someone approaching who has not yet touched the fence. Comparisons of the technologies are in perimeter detection technologies compared.
Privacy
Perimeter cameras often see beyond the property line, onto roads, sidewalks and neighbouring land. In Canada, private-sector organisations are covered by PIPEDA or, in Alberta and British Columbia, by those provinces' own private-sector privacy acts, and public bodies have their own legislation. The common principles are to collect only what you need for a stated purpose, give notice through signage, limit who can view and how long video is kept, and mask areas you have no reason to watch. Analytics zones and privacy masks can usually be set separately. See video surveillance and privacy in Canada for more, and take legal advice for your own situation.
Commissioning and testing
- Set the views. Confirm each camera's view, focus and mounting, by day and by night.
- Draw zones and rules. Set detection zones, lines, masks, object classes and schedules, and record them.
- Walk test. Have people walk, run, crawl and approach at angles through each zone, by day and by night, and at the far edge of the zone. Record which attempts were detected.
- Soak period. Run the system for a period while operators log every alarm and its cause.
- Tune. Adjust zones and settings based on the log, then retest detection to make sure the tuning did not create gaps.
- Repeat seasonally. Retest after the first snow and when vegetation changes.
A typical case
A logistics yard adds analytics to its existing perimeter cameras and is soon receiving dozens of alarms a night. The log shows three causes: headlights sweeping across the fence from a neighbouring road, a camera on a tall, thin pole that moves in the wind, and insects around the built-in infrared lights. The zones are redrawn to stop short of the road, the pole camera is moved to the building wall, and two thermal cameras replace the worst-affected visible cameras along the back fence. After retesting, walk tests are still detected at the far end of every zone and the nightly alarm count drops to a level the monitoring centre can verify one by one.
Questions to ask your integrator
- What detection range does each camera achieve for a person, and how was it tested?
- Where do the analytics run, and what happens to detection if the network or server is down?
- How are events and clips delivered to whoever monitors the site?
- Who tunes the rules after handover, and how are changes recorded?
How ehZee can help
We design camera layouts for perimeter detection and assessment, set out zones and rules, and oversee commissioning, walk testing and tuning. Design drawings are sealed by a licensed professional engineer where required, and installation is by licensed electrical contractors and qualified security installers. See our perimeter intrusion detection service.
Common questions
Can we add analytics to our existing cameras?
Sometimes. It depends on the cameras' resolution, image quality, mounting and views, and on whether the analytics software supports them. Cameras placed for general surveillance often have views that do not suit perimeter detection, so a review comes first.
Will analytics eliminate false alarms?
No system eliminates them. Good analytics on well-designed views, with careful rules and regular tuning, can bring nuisance alarms down to a level operators can manage, which is the realistic goal.
Is thermal always better for perimeters?
Not always. Thermal cameras are strong for detection, especially at night, but cannot identify people or vehicles. Many sites use thermal for detection and visible cameras for identification.