Introduction
Correct deployment of your cameras is crucial to getting the best results from EdgeVis and your chosen VMS system. You want to setup your cameras to maximize correct object detection and minimize false positive alarms.
In order to be effective SafeZone 2D must be used and installed correctly. Here are some things to remember about how SafeZone 2D works:
An object must be visible for at least 2 seconds in the scene
Works best with fixed cameras (with some small vibrations tolerated)
Use with a PTZ camera is possible, but be aware that once the camera stops moving it can take a few seconds for the analytics to learn the new background, and any object already in the scene will likely be treated as background
There must be suitable contrast between background and object in question (see next page)
To detect an object it must be:
taller than 10% of the height of the video frame
AND no taller than 60% of the height of the video frame
AND no bigger than 35% of the area of the video frame
There is no classification of objects – there is only a ‘Detection Start’ and ‘Detection Stop’ trigger available.
How do I make SafeZone 2D more effective?
There are three areas and a number of things you should concentrate on:
Contrast - is it easy to distinguish foreground objects from the background at all times? This includes, optical cameras, cameras using infrared(IR) lighting and detection, thermal cameras. You also need to consider will the surveillance be 24 hours a day? Images with high contrast between the foreground and the background give much better results on correct object detection.
The environment - is there a large amount of moving objects in the scene, or is there hardly any presence of people, cars and trucks? Scenes with minimal presence are much easier to process and generate true alerts of significance.
Maximum object size - is it likely that any people in the scene will take up a large percentage of the image height? Are any vehicles in the scene likely to occupy a large percentage of the surface area? Images where people are likely to be less than 60% of the image height and vehicles that are likely to be less than 35% of the surface area give the best results in terms of correct object detection.
Contrast
The contrast between objects and the background is key to maximizing the effectiveness of object detection within the video streams.
Optical cameras not using infrared should have sufficient artificial lighting
We recommend at least 50 lux of lighting that covers the whole of the detection area when using a day and night IP camera with artificial lighting.Optical cameras with IR lighting
IR spots are good, but false alerts may occur in detection zones where headlight effects are liable to be an issue, especially with cameras that have built-in IR. Any IR lighting should cover the detection area and make it easy to distinguish foreground objects from the background.
The following picture shows an IR camera where the infrared lighting and the camera have been well placed.
The next picture shows a camera where the position and infrared lighting are not well placed. Here the figure towards the back of the image is barely distinguishable from the background.
Thermal cameras should be placed in areas to maximize contrast
These are the best options in terms of performance and detection range for 24/7 surveillance, and they are not subject to false alarms due to headlight effects.
In the following picture the person is easily distinguishable from the background, because the heat level of their body is well in excess of the background temperature.
The next picture shows a thermal image that will be difficult for SafeZone 2D to distinguish the person from the background. Here the background is too hot to make detection easy.
Best camera deployment
Cameras are best placed where the maximum object size is less than 60% of the image height and where the contrast is good. This makes it easier for SafeZone 2D to do its work.
In addition, to aid automatic object detection, it's best to place cameras in significant zones where there is not normally a large amount of traffic, these are known as sterile zones. Due to the fact that there is lower traffic in this type of zone, when an object is detected it is likely to be of significance.
Best performance
Camera is mounted in a location where objects on the stage are infrequent, will be less than 60% of the image height and where the contrast is good.
The next two pictures show cameras located in sterile zones, where no objects or infrequent objects should be observed. The object sizes will never be greater than 60% of the image height and the contrast of the image is good, allowing objects to be easily distinguished.
Moderate performance
It's not always possible to locate your cameras in sterile zones that are also of significant interest. Sometimes you need to compromise to be able to monitor places of real interest. Still, wherever possible you need to try to maximise SafeZone 2Ds chances of making a good identification. In this case you can go for scenes with low levels of activity or possible obstructions, while still making sure that people in the scene will not be larger than 60% of the image height and where any moving vehicles will not cover more than 35% of the image surface. In all cases you should make sure that the contrast between any objects going across the stage and the background is good.
In this work area there is low level activity and a possible obstruction in the large static container on the left but the contrast is good and the rest of the scene satisfies the people and vehicle surface recommendations.
This shopping car park video shows good contrast and each vehicle surface is less than 35% of the image surface. Any people shown will also be far less than 60% of the image height.
Though there is likely to be some traffic within this warehouse delivery bay video, the camera has been positioned so that the trucks will not dominate the scene and people will be under the recommended maximum percentage of height. There is also good contrast in the scene, though the lighting at night would need to be checked.
Examples of poor deployments
Wherever possible you should avoid deployments where:
There are possible obstructions
Vehicles on the stage cover more than 35% of image area
Urban areas where there is significant human activity
Stages where people may be larger than 60% of the image height
Stages where there is insufficient contrast
In this car park scene, the cars represent possible obstructions to detecting people and the nearer vehicles are likely to be larger than 35% of the image area.
This river bridge thermal image shows an area where there is generally a large amount of human activity, but the size of the people in the image and the contrast are good.
Although this house surveillance video is deployed in a sterile zone and the contrast is good, the camera is not located in the best place for automatic detection, as a likely entry route results in images of a person that is over 60% of the image height.
This surveillance video is in a sterile zone and the people in the scene would be at the right scale. However, the contrast is so bad that it's hard to tell if there are people in the scene or not.
Conclusion
The most effective thing you can do to improve object detection quality is to concentrate on the quality of your video feeds. This article has provided examples of good, high contrast, well lit, well staged video streams where monitoring is on a sterile zone. We've also shown good compromises if monitoring a sterile zone is not practical. Finally we've shown examples to avoid if you want to maximize good object detection and minimize false positives.