Axis Digital Autotracking Tested

As camera resolutions continues to climb, the likelihood that you will ever display any camera at full resolution on a monitor declines. This is even more improbable for the normal configuration of 4, 9 or 16 cameras on screen.

How do you help operators handle this?

Axis has a free app for that.

This video shows Axis Digital Autotracking in action. Notice the 2 streams (full and autotracking) and the tradeoff between each one:

Key Findings

Here are our key findings from this test:

Nuisance Activations

• Using default settings with no exclusion zones, digital autotracking frequently "tracked" blowing objects such as bushes and trees moving in the wind.
• Excluding foliage/vegetation, very few false activations were triggered.
• Short lived object/small object settings reduced unwanted activations from passing cars, previously triggering tracking.

Detection Performance

• Detection performance of the 1080p M1125 and 4K P1428-E was similar, with range and nuisance alert rates nearly the same, despite the significant difference in resolution.
• Outdoors, in the daytime, subjects were automatically tracked across a ~275' horizontal FOV/~7 PPF (~225' distance using a ~60° angle of view). 
• Indoors in full light, subjects were tracked successfully nearly 100% of the time across a ~90' HFOV/~21 PPF (~80' distance using a ~60° AOV).
• Low light performance was poor, reducing distances significantly due to increased noise and darkness obscuring subjects. 

Integration/Configuration

• Autotracking stream was available in Exacq, Genetec, and Milestone, but not Avigilon Control Center. Integration with other VMSes may vary.
• Configuration options are similar to Axis VMD 3, with settings to reduce activations due to small, short lived, or swaying objects.

Recommendations

Axis' digital autotracking may assist operators performing live monitoring of a system by drawing attention to moving objects they may not otherwise notice, especially when monitoring large numbers of high resolution (1080p and up) cameras. 

Its detection performance is roughly similar to our past tests of Axis VMD 3 or 2.1, with digital tracking added.

However, detection and exception zones should be carefully configured, along with swaying/small object rules to reduce nuisance activations, and nighttime performance may be poor in very low light scenes.

Availability/Compatibility 

Digital Autotracking is available as a free download directly from Axis. It is compatible with many (but not all) Axis models, including the majority of new (<1 year old) cameras.

Digital Autotracking Stream

When enabling Digital Autotracking, the camera is configured with two streams:

1. The full resolution overview stream, named "View Area 1"
2. The digital autotracking stream. 

Both streams are independently configurable, including resolution, compression, CODEC settings, Zipstream, and more, seen here:

Typically, users would record the full stationary overview stream and only use the digital autotracking stream for live viewing, as it contains redundant video which can be seen by digitally zooming the overview stream during playback.

Stream Size

The autotracking stream is configurable from 160x90 to the camera's max resolution, but defaults to ~1/9 the camera's max resolution. So, the 4K (3840x2160) P1428-E uses a 720p default tracking stream (1280x720), while the 1080p M1125 or Q1615 use a 640x360 tracking stream. 

Based on our tests, we recommend using this default stream size. When using higher resolutions, the digital autotracking "zone" does not center as tightly on objects, but no additional detail is delivered. Lower resolution streams make tracking more difficult, as object size may become too large to be contained in the digitally zoomed region.

VMS Integration

In our tests, Exacq, Genetec, and Milestone were able to utilize the autotracking stream, but Avigilon Control Center was not. 

Cameras using Digital Autotracking add and appear differently, depending on the VMS:

ExacqVision

In Exacq, the stream appears as a secondary stream in their multistreaming options, seen below. Users simply add stream and it appears as a separately configurable stream.

Genetec Security Center

Security Center treats the camera as two separate cameras on a single device, similar to a multi-channel encoder. They may be configured, recorded, and added to views independently.

Milestone XProtect

Devices with Digital Autotracking installed must be added to XProtect using the "Axis 8-Channel Device" driver. The main stream is camera #1, and the autotracking stream is camera #8. Cameras 2-7 are not used and can be disabled.

False Activations

Similar to Axis' other analytics (VMD 3 and 2.1), which Digital Autotracking is based on, false activations from vegetation, shadows, and other nuisance sources are common if using a full screen detection zone without exclusions. 

This can be seen in the foreground of the detection zone setup image below, with the moving grass in the foreground causing near constant alerts/digital tracking.

As with VMD 3, removing these objects from the FOV with exception zones or adjusting swaying object settings eliminates nearly all of these nuisance alarms.

Outdoor Range 

Outdoors, Digital Autotracking was able to track subjects across a ~275' horizontal FOV, ~7 PPF using a 1080p camera. This results in range of about 225' to target when using a ~60 degree angle of view.

The video below shows autotracking using the M1125 on the left, with the full stream on the right. The subject begins walking at ~300' from cameras.

Tracking performance in daytime scenes was similar between the 1080p M1125 and 4K P1428-E. Axis informs us that the P1428-E does not use the full resolution stream for tracking analysis, but a lower resolution stream due to processing limitations. Because of this, both tracked subjects at approximately the same distance in both our outdoor parking lot and field scenes.

This video shows the 4K and 1080p cameras side by side tracking subjects at ~225' in our parking lot scene:

Indoor

Indoors, tracking was very accurate using a ~90' HFOV, ~20 PPF using a 1080p camera. In the scene below, subjects are tracked accurately at the far end of the FOV, ~80-85' from the camera.

Low Light

However, in low light, tracking performance decreases significantly. Our subject is visible at ~80', but the camera does not track him until ~20' range, about a 75% reduction in distance. This reduction is similar to our findings in our test of VMD 3, as well.