Genetec AutoVu LPR Camera TestedBy Ethan Ace, Published Feb 23, 2015, 12:00am EST
IPVM has done many tests on license plate cameras, including the:
- License Plate Capture Shootout
- Low Cost License Plate Capture Shootout
- MP License Plate Camera Tested
- License Plate Recognition (LPR) Axis App Tested
AutoVu claims high read rates, full integration with other systems, and novel advanced vehicle analytics (such as vehicle, speed, direction of motion, state identification and model identification), all uncommon features.
Inside, we have 20+ minutes of videos. Watch the 1-minute preview / excerpt below:
We tested Genetec's LPR on a busy roadway to see how it performed, covering these topics:
- How many vehicles were missed?
- Were plates read accurately?
- What issues caused misreads?
- Where did the vehicle analytics work and fail?
- Plates from all vehicles passing the camera were captured over the course of our testing of ~1500 reads.
- Greater than 98% of plates were read accurately, with partially obscured letters and stacked characters being the main problems
- Plates were occasionally read 2-3 times on the same vehicle, with differing reads for each, though in all cases at least one read was accurate.
- Few incidences of vehicle lettering other than plates being read, such as Fedex or service vehicles.
- Very fast read speed, <1 second from vehicle passing to read event generated in Security Center.
- Speed analytics were very accurate, within 1-2 miles per hour in our tests, at speeds from 10-50 mph.
- Vehicle make analytics missed reading 70-80% of cars. However, when make was read, it was typically accurate.
- Plate state of origin was read only a handful of times. Plates seen in testing were reflective, but the state name was unclear in almost all cases, due to dust and dirt or partially obscured due to license plate frames and covers.
- Direction of travel analytics accurately determined whether vehicles were approaching or moving away from the camera.
- Analytic information is currently only stored with read information, with no alarms or complete searching possible. Genetec says they will implement these features in an upcoming release.
- No Third Party VMS Integration - You have to use Genetec Security Center for AutoVu LPR.
- Fuzzy matching (using * and ?) wildcards does not work. Isntead, users must type the exact string to search. So, for example, searching ">35" returns no search results, but entering "36 mph" returns all vehicles moving exactly 36 miles per hour.
- Analytic data may not be used to create alarms, e.g., A car approaching a camera in a lane in which it should be moving away.
- Sharp XGA 3, firmware version 22.214.171.124
- Genetec Security Center, version 5.2 SR 9 (5.2.1593.16)
Here are our key findings from this test:
The MSRP of Genetec's Sharp camera and included accessories totals ~$7,000, plus another ~$2,000 in licensing costs for AutoVu server, totaling about $9,000 for a single camera system.
AutoVu provides high accuracy, Security Center integration, and novel (though limited) analytic features, though its high price pushes it beyond many users' budgets. However, users looking for top performance in critical applications such as city surveillance, critical infrastructure, or high-end corporate systems should consider AutoVu.
Finally, those seeking VMS integration for platforms than Genetec's own Security Center should look elsewhere. The AutoVu Sharp camera may not be used with other VMSes, and there are currently no third-party VMS integrations of LPR data.
The Sharp XGA is a dual-imager dedicated capture camera. The camera has only a single multipin connector, and no parts are user serviceable.
This video reviews the physical features of the Sharp:
The Sharp camera requires very little configuration when added to a Security Center system. The onboard web interface does not include typical options, such as exposure, CODEC, or other settings commonly seen on standard cameras. Instead, it is mainly used for network addressing and analytic setup such as plate context, vehicle speed, and advanced analytics.
This video reviews the web portal of the Sharp camera:
As part of Genetec's Security Center suite, AutoVu is fully integrated into the Security Desk client, for both live and archive use. Reads and hits appear as events, along with video, alongside cameras and access control devices, and may be searched in the same way.
This video reviews the operation of AutoVu via the Security Desk client:
Camera Field of View
Note that the two cameras of the AutoVu Sharp (LPR and overview) have different resolutions and fields of view. For example, in our testing, we used two cameras, one with a 16mm lens and one with a 25mm lens for the LPR camera. Both used wider angle 8mm lenses for the overview camera.
This image shows the field of view of each:
Mounting and Read Range
Genetec specifies the Sharp may be mounted up to 30° vertically and 50° horizontally from the target at maximum. However, they recommend angles be kept as shallow as possible. This may require moving the camera further back from the read area and using a longer lens option (8, 12, 16, 25, 35, and 50mm lenses are available).
The distance to target varies depending on which lens is used, from about 20' in wider 8mm options to about 100' using 50mm. The read area for the two cameras we used during testing (16 and 25mm) is shown below:
Overall, AutoVu was highly accurate, with few misreads. However, stacked characters regularly caused misreads, as seen in the image below. Most commonly, these characters were read as a single incorrect letter or number, as seen in the lefthand example below. The stacked characters in this case (WR) were read as an M (note that the owl on the license plate was also read).
In the example on the right, the stacked US characters are simply not read at all. This is preferable to characters read wrong, however, as they may still be added to hotlists, either using only the unstacked characters, or using wildcard characters (* and ?).
A second, less common issue was reading vehicle lettering as a license plate. Genetec explains that in some cases where characters are approximately the same height as plate characters, and at the right height in the FOV, they may be read as a plate.
Finally, and least common, were cases where characters were simply not read properly. This is shown in the example below, and likely due to excessive dirt or debris on the plate, effectively "smudging" the letters in the camera's view.
In their latest updates, Genetec released "Advanced Vehicle Analytics", including speed, direction, vehicle make, and state of origin. These analytics are added as annotation fields to each read, and may be displayed in live and search read results.
However, in Security Center 5.2, how these analytics may be used and searched is limited.
Genetec tells us search and alarms will be included in their upcoming 5.3 release.
This video reviews all of these analytics:
Speed and Direction
AutoVu's speed analytics were accurate to within 1-2 miles per hour in our tests once properly calibrated. However, since the camera calibrates speed using the relative size of the license plate, small changes in field of view can drastically change speed detection accuracy.
For example, in the righthand image below, the camera has been angled down and to the right after calibration, a total of only a few degrees. However, this small change results in the ~40 mph truck being read 127 mph.
Relative motion analytics were also accurate, simply reflecting whether the car was moving toward or away from the camera. Since the field of view of the LPR camera is tight and at a shallow angle to the road, only these two directions are determined, and only on vehicles which register a read. Other objects in the field of view of either camera are not analyzed.
In our tests, license plate state of origin was read only one time, and only in our interior lab setting, not field testing. Genetec informs us that the state of origin portion of the license plate must be IR reflective in order to read. Despite PA/NY/NJ being on their list of reflective plates, the state was still not read, in most cases simply a blur as seen below.
Vehicle make was also inaccurate, read in 30-40% of vehicles.
Manufacturer emblems (such as Honda, Toyota, Mazda, Ford, etc.) were more commonly read than those which simply used lettering (Jeep, Volvo).
This test used the following hardware and software versions:
13 reports cite this report:
Back to Top