Realities of License Plate Recognition (LPR)

LPR is a very demanding application that can only succeed in limited operational conditions deployed by expert security integrators.

Historically, publicly available information clearly explaining the operational impact has been hard to find. Thankfully, Milestone has released their LPR administrator's manual providing an honest, clear and concise explanation. I recommend you read pages 29-35 to get a very rapid but deep review of the key factors. Though this is for Milestone the points are generally consistent with the state of the art in currently available commercial systems.

The Milestone document helps to reveal 3 key practical elements:

• LPR can only succeed when a number of strict operational conditions are met.
  
• The costs of achieving these conditions makes LPR unfeasible for many scenarios.
  
• You need deep security integration expertise to succeed but only modest IT depth.

The Conditions

Here are the key conditions that need to be meet in approximate order of difficulty:

• US license plates need to be at least 130 pixels wide. This translates roughly into an image no wider than 5-6 feet assuming 4CIF standard definition video. That's a very tight shot.
  
• The horizontal angle between the camera and plate is within 20 degrees. This means that if your camera is 10 feet away from the plate, the plate cannot be more than 3 feet to the right or left of the camera. This significantly limits where you can put the camera.
  
• The vertical angle between the camera and plate is within 30 degrees. This means that if your camera is 10 feet away from the plate and the plate is 3 feet off the ground, the camera cannot be mounted than 8 feet high. This usually can be accommodated but is low relative to normal heights for outdoor surveillance.
  
• There are a host of lighting adjustments that need to be made. Simply using a stock camera with stock settings will routinely cause very poor performance. For example, Milestone recommends CMOS cameras, disabling auto gain, using WDR and higher shutter speeds (if the car is moving). There is a lot of advanced details that need to be set correctly.
  
• You must use MJPEG and you cannot use H.264 or MPEG-4. Since the analytics in this design are being done outside of the camera and since the analytic can only process images, MJPEG is required. You could theoretically use H.264 or MPEG-4 but then you would have to decode it and the processing power can be very significant. Bottom line is this can have a big impact on bandwidth utilization especially if you are looking for a wireless system.

Feasibility

Clearly, LPR is feasible for the traditional license plate camera use case: A camera installed immediately adjacent to an entrance or toll booth that is only a few feet off the ground and dedicated to looking at the plate. Automated LPR makes reading these plates easier.

However, for broader market usage, this has major limitations. Lots of companies like the concept of monitoring the license plates of people who enter their premises. Setting up cameras in the specific constraints required can be very expensive. Assuming you can find a location that meets these constraints, it requires a construction project that can be $5,000 or more per camera simply for the installation and equipment.

The holy grail is reutilizing your PTZs mounted on roofs and poles. However, these conditions should make it clear that is not feasible. One, getting the resolution needed would be difficult. Does a monitor manually zoom in on license plates? Even if he does, what will the image quality be, given the lighting constraints required for LPR. Also, it will be extremely tough to stay within booth the horizontal and vertical angle requirements.

LPR analysis, with its current capabilities, cannot enable significantly new operational uses of license plate monitoring. While it should help with the traditional use case of monitoring controlled traffic flow, its constraints make it very challenging for broader use.

Security Integration Expertise

The other interesting element that the Milestone manual demonstrates is that LPR integration does not demand deep IT skill but it does demand deep expertise in security design and camera systems.

Integrating LPR is much more like using a graphics design application than it is like setting up a mail server. It depends on understanding the design objectives of security, the physical conditions of the site and the capabilities of the video tools available. The IT elements of the setup are fairly straightforward for a security integrator. The challenge lies in the design and application.

Finally, it is great that Milestone released this manual. Milestone has clearly shared operational limitations that might stop some from buying their product. It is hard for most organizations to do this. Nevertheless, in the long run, it is better for our customers and I believe better for Milestone. In this way, we can maximize the probability that projects will be successful, customers will be happy and the market expands over time.