Testing Avigilon / VideoIQ Rialto
After years of supplying their analytics only on their own cameras, VideoIQ released the Rialto series in 2013, allowing third party cameras to be used. We tested the Rialto with 7 cameras from major manufacturers Axis, Bosch, Sony, and Avigilon (who recently acquired VideoIQ), and put them up against VideoIQ's own HD dome, the iCVR.
We tested in two scenes, an open field, typical of many perimeter protection scenarios:
And a parking lot/driveway, monitoring vehicles in addition to human targets:
In order to answer the following questions:
- How did VideoIQ's own HD dome stack up against the Rialto with third party cameras?
- Which manufacturer performed best?
- What differences were there between 720p and 1080p?
- Were there any compatibility problems?
Here are our key findings from this test:
- Detection of human targets was best when subjects moved across the field of view, instead of toward the camera. Subjects moving away from the camera were tracked consistently farther than the camera could detect subjects moving in.
- False activations were very low in our test scenes, and nearly nonexistent once the cameras' initial learning period was over. Neither moving lights nor shadows, two of the most common sources in analytics systems, caused false triggers.
- Headlights aimed toward the camera obscured vehicle details and caused high missed activation rates (75%+). Headlights moving across the FOV were not a problem.
- Virtual tripwires (IQBeam) are not recommended when using VideoIQ. Object detection rules may be used for the same purpose, including directionality, with more accurate results. IQBeams require more specific setup and area on both sides of the tripwire to function accurately.
- Out of 14 manufacturers we attempted to connect to the Rialto, only one (Samsung) would not connect.
- Diagnostic information for streaming errors is sparse (unsupported resolution, stream not located, etc.), with little indication as to why cameras do not connect.
- Occasionally, we had issues with cameras not streaming at full resolution, despite changing and applying settings. Rebooting the Rialto typically fixed this.
- VideoIQ's own HD dome, the iCVR HD, outperformed most third party cameras connected to the Rialto in the daytime. At night, performance dropped, with most third party cameras outperforming the iCVR due to better low light performance.
- Among third party cameras, Bosch performed best, outperforming both the iCVR HD and other third party cameras, both day and night.
- In full light, 1080p easily bested 720p resolution for range and accuracy by 100' or more. However, low light optimized 720p cameras, such as the Axis P3354, Bosch NBN-733V, and Sony SNC-VB600 outperformed all 1080p cameras in low light.
The Rialto I4 has a street price of about $1,750 USD, while the iCVR-HD sells for about $1,900.
Assuming a third-party outdoor 1080p camera (such as Bosch, Axis, Sony, etc.) sells for ~$100, this pricing makes the Rialto more economical starting at 2 channels ($3,750 for Rialto + cameras vs. $3,800 for two iCVR-HDs). Using lower cost cameras furthers this advantage.
The Rialto is also available in a rackmount R version capable of handling up to 8 HD channels in 1RU. Compared to VideoIQ's own HD cameras, the Rialto R is expensive at small channel counts. Again assuming a third-party outdoor 1080p camera costs ~$1,000, and an iCVR-HD sells for ~$1,900, the Rialto is more expensive until about six channels, when costs are about equal.
In systems with existing cameras, the Rialto may still be preferable, depending on where cameras are mounted, as labor costs to remove and replace cameras could make up this difference.
Overall, third part cameras worked as well as, or better than VideoIQ's own HD model. This may make it especially attractive to in existing systems where switching to VideoIQ cameras when analytics were desired was cost prohibitive. Additionally, for those integrators who prefer to keep working with their preferred camera line, the Rialto is a good choice.
This brief video reviews the physical construction of the Rialto I4:
In this video, we review the process for adding cameras, and provide a brief overview of configuring rules and regions of interest:
Detection of humans performed best when the subject crossed the field of view, as VideoIQ's classifiers could better detect the object's movement as a person.
With the subject moving toward the camera, we can see that VideoIQ's confidence in the object classification is much lower (15% vs. 42% above), despite the subject being closer to the camera.
Headlights moving toward the camera caused significant difficulty in detecting vehicles. Regardless of distance, no cameras triggered on this car moving directly toward the camera.
Taillights, though still very bright, did not cause the same issue.
When using the Axis P3354 with the Rialto, visible noise was frequently seen as objects, shown by the yellow boxes in the image below. Other cameras did not cause this issue. Note that these detections also did not trigger alarms as they were very fast and small, not crossing the alarm threshold.
Outdoor Range Comparisons
We tested the Rialto and cameras in two scenes: an open field with vegetation on multiple sides, ~600' long, to test with human subjects, and a parking lot, including driveway entrance, ~430' long, to test both human and vehicle performance.
This overhead view shows the approximate distances cited in this test:
First, we tested in full daylight, as seen in this FOV:
Camera performance varied, from 250' using Axis and Sony to 375' with the Bosch NBN-932V.
Next, we tested all cameras at night, with light levels dropping to 1 lux and below:
Detection performance was greatly reduced in all cameras, even super low light models such as the Bosch NBN-733V and Axis P3354. The scene is simply too dark with contrast too low to classify objects, varying from about 5 lux in the near FOV to ~0.05 lux at the far end of the field.
Distances are shown on this aerial view:
This field of view shot shows the scene:
Detection of our human subject varied from 250 to 350 feet. Detection of 1080p cameras was generally better than 720p by 50-100', but this scene was simply more difficult, with less contrast between our subject and the background than we saw in our field scene.
All cameras were able to reliably detect vehicles to the bottom of the parking lot. 1080p models were able to detect vehicle passing on the nearby road, 450'+ away.
Finally, we tested all cameras in this scene at night, as well. Light levels were around 1 - 5 lux due to the building's outdoor lighting.
Performance again decreased significantly. Detection of vehicles was possible only within the parking lot, not the road. Detection of our human subject dropped to 100-250'. In this case, low light cameras such as the Axis P3354, Bosch NBN-733V, and Sony VB600 outperformed other cameras by 100'+.
All devices were upgraded to the latest firmware and factory defaulted prior to testing. Cameras were standardized at 1/30s shutter speed, with other settings left default. Current firmware versions:
- VideoIQ Rialto I4: 22.214.171.124094
- VideoIQ iCVR HD: 126.96.36.199091
- Avigilon H3 1.0 MP: 188.8.131.52
- Avigilon H3 3.0 MP: 184.108.40.206
- Axis P3354: 5.40.17
- Bosch NBN-733V: 5.90
- Bosch NBN-932V: 5.90
- Sony SNC-VB600: 1.9.2
- Sony SNC-VB630: 1.9.2
VideoIQ View version 220.127.116.118 was used.