Parking Garage Shootout

BE
Benros Emata
Published May 14, 2011 04:00 AM

Parking garages are a key area for using surveillance cameras. Thefts, assaults, vandalism and accidents are among the many incidents that cameras can help solve.

Monitoring parking garages effectively is more challenging than it first appears. On the plus side, garages are generally enclosed and laid out in long lanes conducive to surveillance camears. However, most parking garages have openings on the side that expose the garages to sunlight. Also, garages depend on artificial lighting that can create great variation of lighting across the scene.

Below is an image from our testing scene showing the presence of outdoor opening and artificial lighting. The cameras were mounted at ceiling level aimed directly down the middle of the driving lane to capture people and cars moving through the garage:

In the test, we used a kit of 7 different cameras (including SD, 720p, 2MP, 3MP and 5MP from Arecont Vision, Axis, Bosch, IQinVision and Sony) to better understand the tradeoffs for ourselves.

The overhead image and map below demonstrates the coverage, distances and pixels per foot at different positions:

 

Then we analyzed a series of scenes to see how performance varied across the cameras. We digitally zoomed each camera to the same level to see the differences. See an example scene comparison from our parking lot test.

We conducted our analysis across 16 scenes. They are:

  • Midday Back Lighting (Far Vehicle)
  • Midday Back Lighting (Near Vehicle)
  • Midday Even Lighting (Far Human)
  • Midday Even Lighting (Near Human)
  • Twilight Back Lighting (Far Vehicle)
  • Twilight Back Lighting (Near Vehicle)
  • Twilight Back Lighting (Far Human)
  • Twilight Back Lighting (Near Human)
  • Nighttime Color (Far Vehicle)
  • Nighttime Color (Near Vehicle)
  • Nighttime Color (Far Human)
  • Nighttime Color (Near Human)
  • Nighttime BW (Far Vehicle)
  • Nighttime BW (Near Vehicle)
  • Nighttime BW (Far Human)
  • Nighttime BW (Near Human)

In many scenes inside the parking garage megapixel delivered minimal advantages compared to standard definition cameras. In some scenes, with relatively even lighting, megapixel delivered modest advantages.

Overwhelmingly, the key challenge was the wide variation of lighting in the scene. For instance, during the day, 20x variation in lighting between the lowest and highest light levels were common in the scene. This variation was even greater than the challenging entrance scene used in our WDR shootout (which had a 13x variation between dark and light areas). At night, the variations reduced but the staggering of artificial lighting created challenges as well. Directly underneath the lights, the lux level rose to approximately 50 lux but in between lights, the lux level dropped as low as 5 lux - a significant spread, even if not as challenging as the day time. 

The challenges demonstrated in our results would be minimixed under certain conditions:

  • Cloudy / overcast days could reduce lighting variation and challenges by half or more
  • Garages facing away from the sun or obscured by trees or other obstructions would have significantly less challenges
  • Artificial lighting with more even light spread would improve night time performance
This noted, we recommend the following:

 

  • Carefully consider and factor in the lighting challenges in parking lots. This can easily be accomplished by bringing a lux meter to the site survey. Changing the physical environment is obviously unlikely but setting the appropriate expectations for challenges can be quite valuable.
  • Do not recommend megapixel as a camera reduction solution if the garage has lighting challenges. This will not work and will create justifiable disatisfaction for the customer.
  • Consider using color over b&w mode at night. Given the artificial lighting, the light levels are not extremely challenging. More importantly, most camera's b&w mode tend to increase glare from artificial light sources (such as car headlights and facility lighting).

Midday Back Lighting (Far Vehicle)

In this scenario, cameras are digitally zoomed into to a vehicle ~60ft out, corresponding to a HFoV of ~56ft. There is ~40,000 lux of back lighting and ~200 lux of illumination in the darker foreground area. As a result lighting conditions are a considerable challenge.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

None of the cameras provide sufficient details for plate capture. However, the higher MP (5MP and 3MP) and 720p cameras reveal fairly crisp contours around the grill pattern and ground effects. As such identification of make, model and color are considerably likely. While not as crisp, the SD cameras provide at least some 'clue' as to the make, model and color. Note that the 2MP color-only is quite washed-out, but still manages to provide some key details (grill pattern and ground effects).

Midday Back Lighting (Near Vehicle)

In this scenario, cameras are digitally zoomed into to a vehicle ~20ft out, corresponding to a HFoV of ~19ft. There is ~40,000 lux of back lighting and ~200 lux of illumination in the darker foreground area. As a result lighting conditions are a considerable challenge.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

All higher MP and 720p provide details for possible plate capture, except for the 3MP (AV3135) which is too under-exposed. Of the MP cameras the 720p (Sony CH140) provides greatest probability of a successful plate capture - with near absolute certainty. Due to lack of details the SD cameras provide no reasonable chance of plate capture and their appreciable level of under-exposure makes identifying make and model less likely as well.

Midday Even Lighting (Far Human)

In this scenario, cameras are digitally zoomed into to a human subject ~100ft out, corresponding to a HFoV of ~93ft. There is ~40,000 lux of back-light and ~200 lux of illumination in the darker foreground area. As a result lighting conditions are a considerable challenge.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

Harsh lighting poses a problem in the higher MP (5MP, 3MP, and 2MP) where the human subject is very washed-out, obscuring facial details. However, a fair sense of the subject's build, gender and hair color/style can be obtained. The 720p cameras provide the same sense of build, gender and hair color/style, but adds a clue as to complexion as well. Note that the 5MP, 3MP, and both 720p provide an indication of a logo on the upper left portion of the subject's shirt.

Midday Even Lighting (Near Human)

In this scenario, cameras are digitally zoomed into to a human subject ~20ft out, corresponding to a HFoV of ~19ft. There is ~40,000 lux of back-light and ~200 lux of illumination in the darker foreground area. As a result lighting conditions are a considerable challenge. 

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

Intense backlighting is causing the subject to be extremely under-exposed here in the darker foreground. Contrast is subsequently low on the subject and details are essentially limited to crude body contours. Under these conditions none of the cameras (MP or SD) can provide identification. Somewhat ironically, compared to the 'far' scenario, overall detail level is diminished - e.g., an ability to detect the complexion and hair color is lost. Note even details such as the presence of a shirt logo is lost as well.

Twilight Back Lighting (Far Vehicle)

In this scenario, cameras are digitally zoomed into to a vehicle ~60ft out, corresponding to a HFoV of ~56ft. There is ~2,000 lux of back lighting and ~30 lux of illumination in the darker foreground area. As a result lighting conditions are a considerable challenge.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

Only the 5MP provides a detail level high enough for a reasonable chance at plate capture. The 720p (Sony CH140) provides the next best chance for plate capture. The grill pattern is fairly crisp in these two cameras as well, improving the odds of make/model identification. Note that the 3MP's especially washed-out image obscures details of license plate, grill pattern, and overall indication of color.

Twilight Back Lighting (Near Vehicle)

In this scenario, cameras are digitally zoomed into to a vehicle ~20ft out, corresponding to a HFoV of ~19ft. There is ~2,000 lux of back lighting and ~30 lux of illumination in the darker foreground area. As a result lighting conditions are a considerable challenge.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

Similar to the 'Midday Back Lighting (Near Vehicle)' composite, all MP (including the 3MP AV3135) and 720p cameras provide a good level of detail and fairly strong odds of a successful plate capture. Additionally, the SD cameras are now providing sufficient details for a slight to moderate chance at plate capture. Overall the 720p (Sony CH140) provides the highest level of details and subsequent odds of plate capture among the composite.

Twilight Back Lighting (Far Human)

In this scenario, cameras are digitally zoomed into to a human subject ~100ft out, corresponding to a HFoV of ~93ft. There is ~2,000 lux of back lighting and ~30 lux of illumination in the darker foreground area. As a result lighting conditions are a considerable challenge.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

None of the cameras provide enough details for identifying the human subject. However, most cameras provide moderate to strong indications of gender, build, clothing, hair style/color, and perhaps complexion. The one exception is the 3MP (AV3135), which is considerably washed-out and obscuring details of the subject's facial details.

Twilight Back Lighting (Near Human)

In this scenario, cameras are digitally zoomed into to a human subject ~20ft out, corresponding to a HFoV of ~19ft. There is ~2,000 lux of back lighting and ~30 lux of illumination in the darker foreground area. As a result lighting conditions are a considerable challenge.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

None of the cameras provide enough facial details for high certainty identification. Indeed, the facial features appear quite blurry or 'smeared' out in the majority of the cameras. However, in contrast to the 'Midday' counterpart there is an overall marked improvement in image quality and detail level. For example, a better sense of the subject's complexion, clothing and hair style/color is given - versus 'Midday Back Lighting' where only crude outlines of the subject were achieved.

Nighttime Color (Far Vehicle)

In this scenario, cameras are digitally zoomed into to a vehicle ~60ft out, corresponding to a HFoV of ~56ft. Note that illumination at the uncovered portion of the parking garage is ~3 lux, and that interior portions are illuminated by artificial lights (~10 to ~50 lux). Note that D/N cameras have been 'forced' to color mode.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

Similar to the 'Twilight Back Lighting (Far Vehicle)' scenario, only the 5MP provides enough details for a reasonable chance at plate capture. Both 720p cameras are the next best as they provide a modicum of definition about the license plate characters. However, the probability of a successful plate capture with the 720p cameras is quite improbable in this particular instance. Note the 3MP is producing significant bloom from head-lights, resulting in an image of rather limited investigative utility - e.g. highly obsured body contours and grill pattern.

Nighttime Color (Near Vehicle)

In this scenario, cameras are digitally zoomed into to a vehicle ~20ft out, corresponding to a HFoV of ~19ft. Note that illumination at the uncovered portion of the parking garage is ~3 lux, and that interior portions are illuminated by artificial lights (~10 to ~50 lux). Note that D/N cameras have been 'forced' to color mode.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

All cameras (both MP and SD) provide a moderate to high level of license plate detail. As such, a moderate to near absolute certainty of plate capture is possible. The 720p (Sony CH140) clearly provides the best detail level of the license plate. Interestingly, the SD cameras exhibit slightly better clarity for plate capture than the 3MP and 2MP cameras, which are somewhat blurry and offer less contrast.

Nighttime Color (Far Human)

In this scenario, cameras are digitally zoomed into to a human subject ~100ft out, corresponding to a HFoV of ~93ft. Note that illumination at the uncovered portion of the parking garage is ~3 lux, and that interior portions are illuminated by artificial lights (~10 to ~50 lux). Note that D/N cameras have been 'forced' to color mode.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

Across the composite, very little detail on the human subject is being delivered. As a result, nowhere is identification a likely outcome. Even where a human is detectable, contrast and clarity is so poor that other relevant information such as gender, build, hair style/color etc. are not readily discernable. Note that on the 2MP and one of the 720p (Bosch NBN-921), it is quite difficult to even detect an actual human subject in the image.

Nighttime Color (Near Human)

In this scenario, cameras are digitally zoomed into to a human subject ~20ft out, corresponding to a HFoV of ~19ft. Note that illumination at the uncovered portion of the parking garage is ~3 lux, and that interior portions are illuminated by artificial lights (~10 to ~50 lux). Note that D/N cameras have been 'forced' to color mode.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

In stark contrast its 'far' counterpart, this composite reveals strong detail levels on the human subject across all cameras, making identification highly likely. Image quality is particulary strong in the 5MP, where the logo on the subject's shirt is quite crisp.

Nighttime BW (Far Vehicle)

In this scenario, cameras are digitally zoomed into to a vehicle ~60ft out, corresponding to a HFoV of ~56ft. Note that illumination at the uncovered portion of the parking garage is ~3 lux, and that interior portions are illuminated by artificial lights (~10 to ~50 lux). Note that D/N cameras have been 'forced' to b/w mode.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

None of the cameras provide sufficient details for a high certainty plate capture. The 720p (NBN-921) is arguably the best in the composite, but provides only a modicum of license plate details and a considerably small chance of plate capture. The bloom from head-lights is quite severe in the 5MP and 3MP, resulting in a loss of useful details that could potentially lead to make, model, and color information.

Nighttime BW (Near Vehicle)

In this scenario, cameras are digitally zoomed into to a vehicle ~20ft out, corresponding to a HFoV of ~19ft. Note that illumination at the uncovered portion of the parking garage is ~3 lux, and that interior portions are illuminated by artificial lights (~10 to ~50 lux). Note that D/N cameras have been 'forced' to b/w mode.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

The majority of cameras provide sufficient details for possible plate capture. The degrees of certainty range from near absolute (3MP and 720p Sony CH140), moderate (720p Bosch NBN-921 and SD Axis M1011) to slight (5MP, 2MP, and SD Bosch NBN-498). While the 5MP and 3MP produce fairly significant levels of bloom from head-lights they are yet able to provide some key details of the grill pattern, ground effects, and overall body contours.

Nighttime BW (Far Human)

In this scenario, cameras are digitally zoomed into to a human subject ~100ft out, corresponding to a HFoV of ~93ft. Note that illumination at the uncovered portion of the parking garage is ~3 lux, and that interior portions are illuminated by artificial lights (~10 to ~50 lux). Note that D/N cameras have been 'forced' to b/w mode.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

None of the cameras provide enough detail for identification. Basically two categories of picture quality emerge from this composite. One in which at best the presence of a human is detectable within the scene (e.g., 5MP and both 720p), and another in which a human subject is not likely to be detected at all (3MP, 2MP, SD NBN-498). Ones in which a human is detectable suffer from such a lack of contrast and details that no other useful information (e.g. complexion, gender, clothing etc.) is likely to be obtained.

Nighttime BW (Near Human)

In this scenario, cameras are digitally zoomed into to a human subject ~20ft out, corresponding to a HFoV of ~19ft. Note that illumination at the uncovered portion of the parking garage is ~3 lux, and that interior portions are illuminated by artificial lights (~10 to ~50 lux). Note that D/N cameras have been 'forced' to b/w mode.

Below is the Composite Snapshot (Download Full Video Clip [link no longer available]):

[link no longer available]

Identifying facial features are produced by all cameras across the composite.

Methodology

The following are the seven (7) cameras used in the 'Parking Garage Shootout':

  • Arecont AV3135 (online $883) - 3MP/1.3MP D/N; 1/2" CMOS; Tamron F1.6; 0.0 Lux (BW)
  • Axis M1011 (online $160) -  4CIF Color-only; 1/4" CMOS; Integrated F2.0; 1.0 Lux (Color)
  • Axis P1347 (online $1350) - 5MP D/N; 1/2.5" CMOS; F1.6 Kowa;0.08 Lux (BW)
  • Bosch NBN-921 [link no longer available] (online $775) - 720p D/N; 1/3" CCD; F1.3 Bosch; 0.4 Lux (BW)
  • Bosch NBN-498 [link no longer available] (online $700) - SD D/N; 1/3" CCD; F1.3 Bosch; 0.06 Lux (BW)
  • IQinVision IQ042SI (online $400) - 2MP Color-only; 1/3" CMOS; F1.6 IQinVision; 0.2 Lux
  • Sony CH140 (online $800) - 720p D/N; 1/3" CMOS; F1.2 Fujinon; 0.1 Lux (BW)

Each of the seven cameras were simultaneously recorded to an Exacqvision VMS during four (4) key scenarios. All cameras are set to their defaults during testing, except during the nighttime tests. During nighttime tests D/N cameras were either 'forced' into B/W or Color mode to ensure consistency of comparison. All camera lenses were adjusted to a uniform lens angle (~50 degrees) and recorded/analyzed at their maximum resolutions.

Here are the key default settings for each camera:

  • Arecont AV3135 - Automatic Exposure
  • Axis M1011 - Exposure Value '50'; Exposure Control 'Automatic'
  • Axis P1347 - Exposure Value '50'; Exposure Control 'Automatic'
  • Bosch NBN-921 - 1/60s shutter; SensUp 4x; AGC on '15'
  • Bosch NBN-498 -  1/60s shutter; SensUp 4x; AGC on '15'
  • IQinVision IQ042SI -1/30 shutter; Color-only
  • Sony CH140 - 1/30s shutter; View-DR/VE On; AGC 'Middle'

Here are the Four (4) key scenarios we examined:

  • Midday Back Lighting - ~40,000 lux (uncovered area); ~200 lux (covered area)
  • Twilight Back Lighting - ~2,000 lux (uncovered area); ~30 lux (covered area)
  • Nighttime B/W Mode - ~3 lux (uncovered area); ~10 to 50 lux (covered area)
  • Nighttime Color Mode - ~3 lux (uncovered area); ~ 10 to 50 lux (covered area)

Each of the four (4) key scenarios feature the same choreographed activity. The activity features a human subject walking towards the camera rig from a 'far' location (~100ft) to  a 'near' location (~20ft). Next, a vehicle subject begins driving in towards the camera rig from a 'far' location (~100ft) to a 'near' location' (~20ft).