Axis VMD3.0 Analytics Vs Avigilon Tested

By Ethan Ace, Published Jan 12, 2015, 12:00am EST (Research)

Axis has released the next generation of their advanced video motion detection application, VMD 3.0. It promises improvements in false alarm rates compared to its predecessor, with new features like swaying object rejection, object size and duration settings, and more.

The question is, how does this new version compare to full-blown analytics like the Avigilon H3A analytic cameras? Can it compete while giving users more options in cameras?

We tested Axis VMD 3.0 running on two cameras, the high end 1080p Q1615 and the 720p integrated IR P3364-LV, against Avigilon's 3.0W-H3A-BO1, to see how they compared in false alerts and coverage area.

Key Findings

We have broken down our key findings into these key areas:

VMD 3 Features

Of the advanced features included in VMD 3, short lived object settings (set to 1-2 seconds) proved most effective at reducing or eliminating false alarms without negative impact on coverage range or missed valid activations.
Small object settings may be useful in some scenes though users should be careful not to set minimum size too small, missing valid partially obscured or distant object.
Swaying object settings reduced alarms caused by moving brush and branches but drastically reduced coverage range both day and night.

False Alarm Issues

Using default settings (no advanced features such as short lived, swaying, or small objects) false activations caused by reflected lights and moving vegetation were most common, with hundreds of false activations per day per camera, both day and night.
After optimizing using short lived object options and exclusion zones, the majority of false alarms were eliminated, reduced to 6-7 false activations per camera per day, similar in both Axis cameras (Q1615 and P3364-LV).
Axis VMD 3 still experienced more false activations after optimization than equivalent resolution Avigilon H3A analytic camera, 6-7 per day vs. 3-4.

Detection Range

Axis VMD 3 effective daytime coverage width using default settings was slightly smaller than equivalent resolution Avigilon H3 analytics camera, ~175' horizontal FOV vs. ~190'.
Effective daytime coverage width with swaying object settings on was reduced from ~175' horizontal to only ~35'.
Nighttime coverage width was reduced to ~30' HFOV in the 1080p Axis Q1615 vs. 60' in the Avigilon H3A cmaera.
Due to its integrated IR, the 720p P3364-LV's nighttime coverage was about 50% better than the Q1615, ~44' vs. 30', despite its lower resolution.

VMS Integration

Axis VMD 3 is integrated only with Axis Camera Station and Genetec Security Center according to Axis. [[Update: Exacq now supports VMD 2.1/3 as an analytic in version 7.4 and later.]]
For VMSes without direct integration, serial data or camera I/O may be used to trigger alerts upon alarm.

VMD 3 Overview and Configuration Recommendations

Axis VMD 3 is a free downloadable camera application compatible with many current and past generation Axis cameras. It performs similarly to VMD 2.1, which we tested previously when using default settings. However, Axis has added three advanced configuration options intended to ignore some common false alarm sources:

Short-lived Objects

This setting ignores objects which are in the scene for a short period of time. We found that setting this to 1-2 seconds eliminated many false alarms seen with it off, such as dust particles or leaves blowing through the scene, or headlight reflections from fast moving cars. Using the 30 second maximum threshold on a defined area may also serve as a rudimentary loitering alarm if subjects are typically passing, but should not be stopping.

Small Objects

Using this setting objects under a minimum size are ignored. This may be useful if detection of human subjects is desired, but small animals should be ignored. It may also be used to ignore blowing dust particles or debris. However, users should be careful to not set minimum object size too small, as this may exclude valid human targets in the back of the scene, or those partially obscured.

Swaying Objects

Finally, sway settings allow users to reduce alarms from objects which repeatedly move only a short distance, such as branches or banners. We found that while turning this setting on did reduce or eliminate alarms from these objects, it greatly reduced coverage range, as much as 50-80% depending on the level selected.

Because of this drastic reduction in range, we do not recommend using this option unless absolutely necessary. It is often possible to exclude regions without swaying objects outright, as many branches may be to the side of the FOV or above the horizon where subjects would not be present. Excluding these regions has no negative impact on coverage in these cases.

Configuration Demonstration

We review these configuration options in the video below:

Test Scene/Procedure

The test scene was selected as it best represented all of the above false alarm issues, with human and vehicle traffic, foliage, and small animals present in a typical commercial application.

We focused first on the most commonly cited causes of false alarms, such as moving vegetation, blowing leaves, small wildlife, and light changes, and secondarily on detection range. Testing was a combination of active testing with human and vehicle subjects, and passive testing, leaving the cameras in place to check for false activations and detection performance.

False Alarm Performance

We tested VMD 3 installed on Axis' high end Q1615 1080p box camera, as well as a P3364-LV with integrated IR. Both were tested alongside an Avigilon 3.0W-H3A-BO1 integrated IR bullet.

Default Settings

Using default settings (no short lived, small, or swaying object options turned on), false alarms were commonly seen from these sources:

Moving branches and brush, both day and night.
Reflected headlights of passing cars at night.

Using these settings we saw no issues from other commonly cited issues such as small animals or blowing leaves.

Optimized Settings

To check its impact on false alarms we set VMD 3's short lived object control to 2 seconds as this matched Avigilon's default settings, and given the size of our test area, a valid object in the field of view would be present for at least this period of time.

We left small object settings off, as we wanted to alarm on distant humans and vehicles which may drop below this threshold, and left swaying object settings off to maintain detection range.

Using these settings, false alerts dropped substantially. Alarms from blowing brush and branches were still most common, though at a drastically reduced rate (5-6 per day vs. 8-10 per hour). The image below highlights problem areas which triggered these alarms.

VMD 3 did not activate on either camera due to leaves blowing through the FOV, circled below. The Avigilon H3A camera triggered on these leaves 3-4 times over the course of testing.

Using optimized settings, VMD 3 was also not affected by reflected lights as Avigilon was at times. The small highlighted area shows the reflection of car headlights from the road nearby, which triggered the H3A camera multiple times over the course of testing. The larger highlighted light occasionally triggered the Avigilon analytics as object passed by, though this did not occur in either camera using VMD 3.

Finally, we tested the cameras against changing light indoors, including opening and closing the door of a dark room, turning lights on and off, and flashlights reflecting on the walls of the scene. None of these issues triggered alerts in the cameras running VMD 3, nor Avigilon.

Coverage Performance

In our test scene, coverage width of the 1080p Axis Q1615 running VMD 3 was only slightly less than the Avigilon 3.0W-H3A-BO1 during the day, ~175' vs. ~190' horizontal field of view. The 720p Axis P3364-LV's coverage was ~155' due to its lower resolution.

At night, coverage using the Q1615 drops to only about a ~30' HFOV due to its lack of integrated IR. The P3364-LV dropped to ~44' HFOV, ~50% better than the Q1615 due to its built in IR. The Avigilon H3A camera bested both, at about a 60' HFOV.

Integration

Axis VMD 3, like 2.1 before it, integrates only with Axis Camera Station and Genetec Security Center out of the box. Other VMSes have two potential options for integrating these events:

TCP Serial Strings

Many VMSes, such as ExacqVision and Milestone XProtect, accept raw serial data over TCP which may be used to drive events. Users create an event in the camera which sends a specific string upon VMD 3 activation. The VMS listens for this string to trigger recording and/or alerts. This process is detailed in this Exacq support document.

The main drawback to this method is time. Creating event rules on each camera and corresponding actions in each camera and confirming they work properly may take 10-15 minutes per camera. This quickly adds up in even mid-sized systems.

Camera I/O Ports

A second option is to configure the camera to trigger a relay output upon VMD 3 alarm. This output is then codnnected to the camera's input. The VMS may then be configured to trigger recording and/or alerts upon activation of the camera input.

Like the serial integration described above, this requires events to be created in both camera and VMS, which can be time-consuming. Additionally, this method assumes the VMS supports the camera's input as an event trigger, and that I/O is not required for any other purpose.

Update: Exacq Support

Exacq supports VMD 3 as an analytic (along with other Axis analytic events) in version 7.4 and later. Users may configure the camera to record on VMD 3 using Event Linking, seen here: