Best practices
IRIS+™ is a cloud-based video analytics Software as a Service (SaaS), powered by cutting-edge Deep Learning technology that enables unparalleled detection accuracy.
IRIS+ supports any fixed camera by connecting any RTSP/ONVIF video stream with the IRIS+ Edge, a compact appliance with pre-loaded software.
Cameras can be IP cameras or analog cameras through a DVR that can output an RTSP/ONVIF stream.
This document describes best practices and considerations for selecting the cameras, viewing angles, ONVIF/RTSP resolutions and frame rates, and various aspects and scenarios of deployments to achieve the best analytics performance.
The integrator must perform a comprehensive Site Survey of the specific site on which the deployment of the IRIS+ video analytics system is planned, with the objective of identifying site parameters that may affect the ability to satisfy both project requirements and video analytics performance requirements.
The following items are documented in the Site Survey:
- Requirements: Detailed video analytics requirements, to be used as acceptance criteria for the deployed solution
- Cameras: Cameras used (with their site layout) or planned to be used with video analytics
- Viewing application of video and alerts: 3rd party VMS / PSIM / alarm monitoring applications will be used
- Deployment architecture: physical location of the recording, viewing, and monitoring clients and servers Forensics / Statistics: if utilized, specifying operator locations, interfaces, Data retrieval mechanism, retention time, and export format
- Network Infrastructure
- Failover requirements
- Lighting conditions
- Environmental factors
This section describes how to select cameras for a site where the cameras are not yet installed or selected. Select from the following camera types:
This is the most common camera type used with video analytics. There is a long list to choose from, ranging from VGA to HD/FHD and higher resolution IP cameras. IRIS+ can also work with analog/HD Analog by retrieving a RTSP/ONVIF stream from a supporting DVR or analog video converter.
Analytics performance can be increased by deploying the capabilities of:
- Wide Dynamic Range
- Automatic switching between day and night
Select a thermal camera in case of either:
- Limited or no lighting is available at night -or-
- You want the detection to cover large distances (over 80 meters) with a single camera
It is not advisable to deploy thermal cameras with rules utilized to detect stationary targets such as:
- Stopped Vehicle
- Suspicious Object
- Asset Protection
- Forensic search based on color
Type | Characteristics | Main Usage | Compatibility with Video Analytics |
|---|---|---|---|
Standard CCD / CMOS | Commonly used color camera | Any common scenario | Very high |
Thermal | Infra-Red radiation imaging (temperature based) | Complete darkness Very long distances | Very high for perimeter protection application, some advanced scenarios not supported due to grayscale thermal image |
Panoramic/ Fisheye | Circular image – usually 180° / 360° | Wide panoramic coverage of an area with a single camera | Currently limited to top-down ceiling mounted, supporting only people-focused analytics such as moving in an area, crossing a line, and counting |
Type | Characteristics | Main Usage | Compatibility with Video Analytics |
|---|---|---|---|
Box/Bullet | The box-shaped body requires a mounting arm or pole | Commonly used Indoor / Outdoor camera | Very high |
Dome | Dome shape, usually with a built-in enclosure. Ceiling, wall, and arm/pole mounting | Commonly used Indoor / Outdoor camera | Very high |
PTZ | Pan-tilt and zoom camera | Guard tour across multiple presets, tracking objects | Currently not supported unless positioned statically and never moved. |
Based on analytics rules requirements, selected camera types, and recommended detection distances, you can next determine camera placement and required fields of view for each camera. See also
It is recommended to use cameras with viewing angles (horizontal FOV) that are narrower than 110°.
Fisheye cameras or cameras with a horizontal field of view that is wider than 110° may suffer from severe image warping that degrades analytics performance.
The camera should be properly leveled to the horizon plane (the horizon should not be tilted).
- To determine the ideal field of view (FOV) for video analytics per camera, the following main criteria should be taken into consideration: Maximum target distance: The maximum distance between the camera and the analyzed target, allowing reliable analytics. Determined by the camera’s viewing angle and the size of targets at maximum distance.
- Minimum target distance: The minimum distance between the camera and the analyzed targets, allowing reliable analytics. Determined by the camera’s viewing angle and the size of targets at minimum distance.
- Target separation: A camera’s field of view is considered to have good target separation if there are minimum occlusions caused by adjacent or close-by objects, allowing a complete view of each target separately. Good target separation is key to the successful deployment of video analytics in scenes that are non-sterile.
- Camera resolution: Irisity’s analytics supports resolutions of 320x240 (recommended only on thermal cameras) and higher including HD and megapixel resolution, although analytics accuracy will not be improved by resolutions higher than FHD.
Each of the above criteria can be altered based on the following guidelines:
Increase Maximum Target Distance | Decrease Minimum Target Distance | Improve Target Separation Option 3 |
|---|---|---|
Increase target magnification by using longer focal lengths | Decrease target magnification by using shorter focal lengths | Increase the tilt angle while striving for a 90-degree angle (camera looking straight down) |
Decrease the camera’s tilt angle* (Note1) to extend the camera’s vertical viewing angle / maximum viewing distance | Increase the camera’s tilt angle*(Note1) to shorten the ‘dead-zone’ | Use higher mounting heights to achieve top-down viewing in wider areas across the field of view |
- | Use greater mounting heights to produce lower-sized targets which do not suppress the maximum allowed target size | - |
Note: Zero degrees tilt angle refers to cameras mounted parallel to the ground. 90-degree tilt angle refers to cameras mounted orthogonally to the ground (looking straight down) that are currently not supported
Conclusion:
- For the long-range field of view, use increased focal length and low tilt angle. The minimum viewing distance can be compensated by adding a single camera at the top of the camera chain.
- For the short-range field of view, use short focal lengths, high tilt angle (as close to 90 degrees as possible), and higher mounting. The higher the camera is, the higher the tilt angle required.
- For non-sterile scenes with the continuous presence of multiple people and/or vehicles, and for crowded environments, optimize target separation by using large tilt angles (90 degrees or as close as possible to it) and higher mounting. Height and tilt angles compensate for each other in this case.
Adequate Field of View (FOV) and Area of Interest (AOI):
- Position the camera so that it mostly occupies the AOI. Exclude irrelevant areas from the FOV such as the sky and other non-important regions. Try to keep the AOI centered in the FOV.
- Avoid concealments and truncations of detected targets as much as possible. The view of the AOI is best without physical obstacles such as trees, buildings, poles, signs, and any other static object that may obstruct the view of the target (person, vehicle, or static object).
- Avoid vegetation, puddles, and running water in the AOI.
The effective range for reliable analysis and detection is determined according to target size criteria which differ from one scenario to another.
- The different scenarios are detailed in the below table: Size range criteria are based on the relative size of targets in the FOV. They are required for reliable analysis of people, vehicles, and static objects. Size range criteria are used across typical scenarios of camera mounting and scene learning.
- ‘angled’ refers to cameras positioned with a tilt angle of 20°-50° (cameras tilted at an angle of 0°-20° or 50°-85° present views that are not optimal for analytics and are therefore not recommended)
See Camera placement principles for more details regarding the Camera Placement Design Tool.
- Best practices for size range for people analysis
Scenario | Camera Orientation* | Minimum Size Criteria | Maximum Size Criteria | Example |
|---|---|---|---|---|
Walking/running / loitering/crowding | Angled | Target height is greater than or equal to 36 pixels in length | Target’s longest side is less than 60% of the parallel image edge | See reference below |
- Size range for vehicle analysis, as a % of the FOV
Scenario | Camera Orientation* | Minimum Size Criteria | Maximum Size Criteria | Example |
|---|---|---|---|---|
Moving target on a single-lane road/street | Angled Vehicle is viewed from its side | Target's longest side is greater than or equal to 36 pixels in length | Target’s longest side is less than 60% of the parallel image edge | See Reference 2.1 below |
Static target on a single-lane road/street | Angled Vehicle is viewed from its side | Target's longest side is greater than or equal to 36 pixels in length | Target’s longest side is less than 60% of the parallel image edge | See Reference 2.1 below |
Crowded scenarios, people or vehicles | Angled | Target's longest side is greater than or equal to 36 pixels in length. Ensure separation between different objects in view. | Target’s longest side is less than 60% of the parallel image edge | See Reference 2.1 below |
Moving target on a multi-lane road/highway | Overhead/Angled mounted on the gantry or angled camera on pole (6 meters or more) | Target's longest side is greater than or equal to 36 pixels in length. Ensure separation between different objects in view. | Target’s longest side is less than 60% of the parallel image edge | See Reference 2.2 below |
Static target on a multi-lane road/highway | Overhead/Angled mounted on the gantry or angled camera on a pole (6 meters or more) | Target's longest side is greater than or equal to 36 pixels in length. Ensure separation between different objects in view. | Target’s longest side is less than 60% of the parallel image edge | See Reference 2.2 below |
Moving target in an open area (parking lot, open yard) | Viewed from multiple directions including front and back with angled cameras | Target's longest side is greater than or equal to 36 pixels in length | Target’s longest side is less than 60% of the parallel image edge | See Reference 2.3 below |
Static target in an open area (parking lot, open yard) | Viewed from multiple directions including front and back with angled cameras | Target's longest side is greater than or equal to 36 pixels in length | Target’s longest side is less than 60% of the parallel image edge | See Reference 2.3 below |
Moving target on a single-lane road / street | Angled Vehicle is viewed from its side | Target's longest side is greater than or equal to 36 pixels in length | Target’s longest side is less than 60% of the parallel image edge | See Reference 2.1 below |
- Size range for static objects analysis, as a % of the FOV
Scenario | Camera Orientation* | Minimum Size Criteria | Maximum Size Criteria | Example |
|---|---|---|---|---|
Low activity scene Good illumination Non-appearance of shadows, reflections, glare | Overhead/Angled | Target's longest side is greater than or equal to 36 pixels in length | Target’s longest side is less than 60% of the parallel image edge | See Reference 3.1 below |
Any indoor/outdoor scenario with good illumination | Overhead/Angled | Target's longest side is greater than or equal to 36 pixels in length | Target’s longest side is less than 60% of the parallel image edge | See Reference 3.2 below |
Reference 1.1
- Walking/running / loitering/crowding with an angled camera.
- A person’s longest side is approximately 36 pixels
- The size is within the allowed range for all person-related video analysis functions.
Reference 2.1
- Vehicle driving/stopping with an angled camera and side view
- The vehicle’s length is approximately 36 pixels.
- The size of the vehicle is within the allowed range for moving/stopped vehicle analysis
Reference 2.2
- Vehicle driving/stopping on multiple lanes with an angled camera and side view
- The vehicle’s length is approximately 36 pixels.
- The size of the vehicle is within the allowed range for moving/stopped vehicle analysis.
Reference 2.3
- Vehicle driving/stopping on an open area with an angled camera and side view
- The vehicle’s length is approximately 36 pixels.
- The size of the vehicle is within the allowed range for moving/stopped vehicle analysis.
Reference 3.1
- Analysis of static objects/bags left behind.
- The length is approximately 36 pixels.
- The object is within the allowed size range for any static object analysis scenario.
- One of the most crucial points to take into consideration in an analytics deployment is the amount of light available during operational times. A high-quality image is critically important for successful video analytics deployments and so light considerations are key to ensuring high-performance video analytics.
- Guidelines are provided below for best practices when deploying analytics under different lighting conditions for achieving the best possible performance. Please disregard this section when using thermal cameras.
- Comprehensive performance testing on site is achieved when you can identify the target you want to detect in all lighting conditions, 24/7.
- Security cameras utilize several key elements to control the amount of incoming light and thus improve image quality in poor lighting conditions. The main elements which impact the amount of incoming light are size of the sensor, iris/aperture, gain level, and shutter speed.
- Auto-iris: When properly implemented, auto-iris maintains correct exposure levels and usually delivers moderate and gradual brightness changes. So, if you need to operate video analytics across varying lighting conditions, choosing cameras that support auto-iris is best.
- Auto-shutter speed adjustment: Since the negative impact of slow shutter speed can always be expected when motion occurs, don’t rely exclusively on automatic shutter speed capabilities to properly compensate for video analytics under poor lighting conditions, since the capability to properly detect motion is essential.
- Automatic Gain Control (AGC) and Noise Reduction Filters (DNR): In practice, the combination of AGC and DNR usually results in rapid contrast changes (aka ‘video breathing’) accompanied by a certain level of the remaining noise. Both are known as negative factors for video analytics as they manifest in rapid pixel changes often interpreted as nuisance motion, which increases the probability of false alarms.
- Avoid manual / fixed customization of the iris, shutter speed, and gain which aims to improve image quality at night, due to the negative impact of such adjustments during daylight hours
- Favor day/night cameras that can adjust to nighttime conditions and light changes utilizing auto-iris functionality, optionally combined with moderate AGC and shutter speed control, over day/night cameras supporting AGC with static or fixed iris control. A good example of this technology is called P-Iris.
After determining the best method to improve nighttime video quality, be aware of the rules of thumb when applying each. Here are the main ones:
- There is a wide range of IR illuminators and price points. In most cases, higher-cost illuminators will provide better image quality, longer distances, and longer lifetime.
- IR suffers from shallow depth of view, meaning that a single illuminator is mainly useful when used for either short range or longer ranges, but not both. If you need to cover a range of more than 10 meters, you may need to combine short and mid-range illuminators for the same camera.
- Every illuminator has a different beam width, allowing it to cover a particular width of the field of view. Wider angles allow shorter distances and vice versa. Make sure to choose an illuminator with an angle of view as close to the camera’s angle of view as possible.
- Use day-night cameras containing an IR cut-filter (which disables IR during daytime or in good lighting conditions) unless the camera is only used in dark scenes.
- Always focus the cameras at night when the IR is in action (IR cut-filter is off), to bypass the known IR focus shift issue.
- IR achieves greater distances indoors than outdoors. IR range indicated by vendors is usually based on indoor scenes. According to performance review indications found in the public domain, when deploying IR outdoors, expect to achieve detection distances that are half to a third of the distance indicated in the IR lighting product specification.
- The illuminator should be mounted at least 1 meter below the camera.
The main criteria of video quality for reliable analytics when utilizing thermal imaging is the contrast level differences between the detected target objects and their surrounding background. Higher contrast differences mean better quality. While the level of contrast difference is directly impacted by the temperature difference between the target object and its surroundings, the actual figures vary in different environmental conditions (temperature, humidity, etc.) and distances.
As a result:
- Pay close attention to the vendor’s specifications regarding achievable distances and how they are impacted by environmental conditions.
- Test the camera performance in the field in varying conditions. See also Testing Lighting Performance.
- Some thermal cameras adopt various techniques to dynamically control their imaging sensitivity in varying conditions. These methods sometimes result in rapid luminance and contrast changes. Since video analytics uses increased sensitivity when deployed with thermal cameras, it may lead to nuisance of false alarms caused by these rapid changes. Consult with the camera vendor to ensure that the camera allows some level of control or disabling of these functions.
Sufficient lighting level is the most common criterion to consider when deploying analytics. There are, however, additional challenges for video analytics concerning the way that light and camera placement impact image quality.
Here are the main issues to consider and how to resolve them:
Scenes containing a wide range of lighting conditions, including extremely bright and dark areas, will generate an image where the dark areas are completely dark, and the bright areas are saturated.
Common scenarios involve indoor scenes with natural light coming from the outside (windows, entrance or garage doors, etc.), or scenes containing shadowed areas.
Resolution:
Use a camera with good WDR functionality or, alternatively, consider using a thermal camera.
Direct light from vehicle headlights combines an extreme WDR situation with low illumination, resulting in an extensive glare around the headlights and complete darkness around it.
Resolution:
Cameras used for video analytics should not be directed toward vehicle headlights. Avoid using the same camera for LPR and video analytics. If headlights reflect blindingly on a road or wet surface:
- Low severity level
Select a camera with advanced WDR function or IR illumination combined with IR bandpass filter lens
- Medium severity level
Use a slow shutter speed combined with IR illumination
- High severity level
Select thermal cameras
Direct sunlight ‘blinds’ the camera and generates ‘sun spots’ caused by reflected light, similar to the human eye when looking towards a strong light.
Resolution:
- Exclude the sky from the camera’s field of view
- Note the course of the sun during the day and position the camera so that it will not face the sun directly at dusk and dawn
- Avoid placing the camera in front of windows, otherwise use curtains or plants to cover the blinds as much as possible
Resolution:
Video analytics can handle moderate shadows and reflections. When dealing with shadows in a controlled, narrow-angled scene such as in deployments of people counting or tailgating where the camera is placed above an entrance or close to a door, consider placing a carpet (preferably grey colored) on the floor to eliminate the shadows. In extreme situations, use thermal cameras.
The table below summarizes the main lens and camera imaging functions and features to be considered and provides observations and recommendations for each.
Function / Feature | Observations and Recommendations |
|---|---|
Ultra-high resolutions (above FHD) | - Higher resolutions do not increase the detection range - Poorer image quality in low light conditions compared to FHD resolution - Improved image quality during daytime WDR scenes |
Auto-Iris / Manual Iris | - Auto-Iris is mandatory for outdoor locations with varying lighting conditions - Manual Iris is a suitable alternative only for constant lighting conditions - Fixed Iris should be avoided |
AGC | Standalone AGC functionality (which is not part of a combined module such as P-Iris or multi-function low-light capability) is usually a negative factor and should be avoided or disabled. |
WDR | Highly recommended for WDR scenes during the daytime, if a high quality ‘true’ WDR is used (aka multi-exposure WDR) |
Advanced extreme low-light techniques | Doesn’t overcome the need for external illumination in low-light conditions under 5 lux, due to its disadvantages. It may, however, eliminate the need for illumination in higher lux levels. |
Integrated IR | Meaningless in most scenarios except for in cases of extremely short ranges such as deployments of people counting or asset protection with an FOV range of up to 4-5 meters. |
IR cut filter | Mandatory when using IR illumination during daytime. |
IR bandpass filter | Recommended when using IR and dealing with glare from ambient lights. |
Varifocal lens | - Mandatory when the FOV and AOI are not strictly determined at the stage of choosing the cameras, and when the FOV and AOI may change in the future. - Highly recommended in any other case since it allows fine-tuning the FOV. |
Interchangeable Lens | - Mandatory when needing to achieve extremely wide or extremely long (zoomed) FOV beyond the capabilities of the default camera lens. - It’s highly recommended to choose cameras with the interchangeable lens when the FOV and AOI are not strictly determined at the stage of choosing the cameras. |
Auto-focus | Highly recommended as it significantly simplifies the focusing process, especially when the cameras are mounted high and/or in locations that are hard to access. |
- Detect intruders approaching or loitering along the perimeter barrier (usually fence line or wall)
- Detect intruders crossing the perimeter barrier
- Person crossing a line
- Person moving in an area
- Person loitering
- Make sure that poles are stable in winds (if you mount cameras on poles) because pole movement decreases analytics performance. Standard electricity poles are insufficient as they sway excessively in wind.
- Calculate the ‘dead zone’ and detection range for the cameras being used, to determine the distance between cameras.
- Point the camera view along the perimeter so that movement of intruders is captured in an orthogonal direction relative to the camera when approaching the perimeter.
- Point the camera so that it views and protects the inner or outer area of the perimeter. Don’t attempt to point the camera so that it views both sides of the perimeter.
- Position the cameras so that each ‘dead zone’ is covered by the adjacent camera. In the following example, the Green colored cameras cover the blind spots of the opposite cameras located in the corners of the perimeter. With this deployment method, there are no dead zones.
When protecting long perimeters, camera counts are a key parameter as they have a significant impact on the overall deployment cost. To minimize the required number of cameras while not compromising on the probability of detection, follow the recommendations in section 8, Optical Camera Recommendations, and section 9, Thermal Camera Recommendations. As a rule of thumb, cameras can often be spaced 100-200 meters or more apart depending on chosen resolution and camera field of view.
Camera Mounting Height and Angle
- Mount cameras at least 3.5 meters / 11.5 feet from the ground.
- Cameras should be mounted with a certain downward tilt / mounting angle to ensure movement to and from the camera creates significant pixel changes. When covering smaller areas, a mounting angle of 20-60 degrees is often a good choice. For longer distances, adjust as necessary.
Cameras with wide viewing angles cover wide visible areas, however, they limit the analytics detection distance (see Table 1 - Estimated Detection Ranges of Optical Cameras or Table 2-Estimated Detection Ranges with thermal cameras). It is recommended to use cameras with viewing angles (horizontal FoV) that are narrower than 100°. Fisheye cameras or cameras with a horizontal field of view that is wider than 100° may suffer from image warping that degrades analytics performance.
- The recommended video stream frame rate is 8 – 12 FPS.
Note:
- Lowerframe rates will degrade analytics performance
- To meet the minimum required frame rate, the camera’s shutter speed should be at least 1/12 seconds
- Higher frame rates consume additional computing resources but will not result in improved analytics performance
- The recommended video stream resolution is 720p – 1080p.
Note:
- Using 1080p video stream resolution will result in improved detection (see the table below) but will consume more computing resources and network bandwidth
- Using video stream resolution that is higher than 1080p will not necessarily result in improved analytics performance as megapixel cameras generally suffer from degraded low light.
- Pixel on target/ pixels per meter/ pixel per foot:
- The system requires 36 pixels on target for most applications, in a non-obscured view
- For a typical person of 1.80m/ 5.9ft height, that means 20px/m or 6.1px/ft
- For a typical vehicle of 5.8m/15.7ft length, that means 7.5px/m or 2.3px/ft
Note: Please also take into consideration that cameras with narrow FOV are more sensitive to camera shake, which could impair detection performance
The following environmental conditions may degrade analytics performance:
- Obscured targets
- Camera shake
- Fog
- Inadequate lighting at the required detection distances
- The recommended minimal frame rate is 8FPS.
Note:
- Lower frame rates will degrade analytics performance
- Higher frame rates consume additional computing resources but will in improved analytics performance
- The recommended resolution is VGA (or above)
- Pixel on target/ pixels per meter/ pixel per foot:
- The system requires 36 pixels on target for most applications, in a non-obscured view
- For a typical person of 1.80m/ 5.9ft height, that means 20px/m or 6.1px/ft
- For a typical vehicle of 5.8m/15.7ft length, that means 7.5px/m or 2.3px/ft
Note: Please also take into consideration that cameras with narrow FOV are more sensitive to camera shake, which could impair detection performance
To minimize the ‘breathing’ effect created by thermal cameras’ AGC, it is recommended to (1) use lenses with small focal lengths and (2) set the camera’s maximum gain to be in the range of 3-9dB.
The following environmental conditions may degrade analytics performance:
- Obscured targets
- Camera shake
E-mail: [email protected]
Phone: +46 771 41 11 00