Intersection Shootout

Author: Benros Emata, Published on May 07, 2011

Intersections are an important area for using surveillance cameras. Accidents, thefts and traffic congestion are among the many common incidents that cameras can help solve.

However, intersections are particularly challenging to monitor. The scenes are complex with many vehicles and people moving through it. The reflective surface of the cars and their headlights at night make it harder for surveillance cameras to accurately capture details.

Megapixel cameras should provide an important advancement in monitoring intersections as they deliver a significant increase in pixel density relative to traditional CCTV cameras. How much of an advancement is it really and under what conditions?

In this report, we share our extensive findings from a real world intersection test. We went out to a busy intersection with a kit of 7 IP cameras (including SD, 720p, 2MP, 3MP and 5MP from Arecont Vision, Axis, Bosch, IQinVision and Sony) to better understand the tradeoffs for ourselves. As the image below shows, we captured most of the intersection from one side to the other:

We placed our camera kit on one corner to capture both directions of traffic. 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 13 scenes. They are:

  • Midday Comparison By Distance / HFoV
  • Night Comparison by Distance / HFoV
  • Midday Intersection - Vehicle (Far)
  • Midday Intersection - Vehicle (Mid)
  • Midday Intersection - Vehicle (Near)
  • Midday Intersection - Person (Far)
  • Midday Intersection - Person (Near)
  • Night Intersection - Vehicle  (Far)
  • Night Intersection - Vehicle (Mid)
  • Night Intersection - Vehicle (Near)
  • Night Intersection - Person (Far)
  • Night Intersection - Person (Near)
  • Midday Half Intersection Analysis

Inside, we dig into and provide recommendations on each one, sharing our findings from the test.

[NOTE: Videos and images are not available for this test. They were deleted during a switchover to a new storage provider. I apologize for the problem].

Key Findings / Recommendations

Get Video Surveillance News In Your Inbox
Get Video Surveillance News In Your Inbox

While the megapixel cameras delivered moderate improvements in details captured, overall the performance was underwhelming, especially relative to the megapixel marketing hype. Certainly benefits exist with megapixel but expectations must be moderated up front.

  • For vehicles, the megapixel cameras made it easier to determine make and model by providing more details on the shape and accessories of the vehicles. On the negative, it was nearly impossible to see details of plates or people inside vehicles.
  • For people, the megapixel cameras provide modest amount of increased details but nowhere could the cameras provide clear facial details (even relatively close to the cameras).
  • Reflection off of vehicles and headlight glare were major problems - more so for the megapixel cameras (and especially bad for the Arecont Vision camera).

When looking to deploy surveillance at an intersection, the following findings from our test should be kept in mind:

  • The dimensions of the intersection: Our test location was 6 lanes by 6 lanes and measured 80 feet by 80 feet (just the blacktop/road, not counting the sidewalk). Setting up a camera to be wide enough to cover this creates challenges in capturing details even for megapixel. In our test setup, we only covered about 2/3rds to 3/4th of the intersection. In a smaller intersection or choosing to cover only half an intersection, visual details captured could improve significantly (see last scenario in this report).
  • In contrast to a parking lot of similar dimensions, the captured video quality at an intersection is likely to be consistently worse. Rapid lighting changes from reflection off of moving vehicles and bursts of headlights are common issues at intersections that constrain megapixel cameras from providing their potential quality improvements.

Day Comparison By Distance / HFoV

The image below overviews the difference in image detail displayed by the camera capturing the most and least details for 2 different locations in the scene during the day. Even fairly close to the camera (with a ~33ft HFoV), while the MP provided more detail, the practical benefit was modest at best. On the other side of the intersection, the megapixel camera provides a bigger 'blob' which could help marginally in describing a suspect. For vehicles, the general benefit was that megapixel made it easier to identify the make of a vehicle, but not the license plate.

Night Comparison By Distance / HFoV

At night, the same patterns seen during the day occur. Interesting to note, because of the heavy presence of vehicle headlights, the image was relatively bright for night time but the headlights caused significant imaging issues for some cameras.

Midday Intersection (Far Vehicle)

In this scenario, cameras are digitally zoomed into vehicles ~130ft out, corresponding to ~121ft HFoV. Midday lighting is even and not particularly challenging.

None of the camera images provide enough details for plate capture. However, in the majority of MP cameras, other key details are elucidated. For example, color, vehicle type (sedan/SUV etc.), and details in the grill pattern help to identify make and model. In contrast, the SD images provide only a crude indication of color, and arguably vehicle type.

Midday Intersection (Mid Vehicle)

In this scenario, cameras are digitally zoomed into a vehicle ~85ft out, corresponding to a ~79ft HFoV. Midday lighting conditions are even and not particularly challenging. Note that one of the 720p (Bosch NBN-921) cameras is not represented in this composite, due to an issue with lower frame-rate.

Again, none of the camera images provide adequate detail levels for plate capture. However, all cameras do provide enough detail to identify that the vehicle type is a compact sedan and that its color is likely to be either a variety of white, grey or lighter shade of blue perhaps. In the MP cameras (notably the 5MP and 2MP) a fair level of detail in the vehicle contours and grill pattern provides a moderate to strong likelihood of make/model identification.

Midday Intersection (Near Vehicle)

In this scenario, cameras are digitally zoomed into a vehicle ~55ft out, corresponding to a ~51ft HFoV. Midday lighting conditions are even and not particularly challenging. Note that one of the 720p (Bosch NBN-921) cameras is not represented in this composite, due to an issue with lower frame-rate.

Even at this 'near' vehicle location, none of the cameras provide detail levels sufficient for successful plate capture. Note that the 5MP camera provides roughly 50ppf in this particular 'shot'. However, overall clarity has been enhanced compared to the 'mid' scenario. And, as such, other key details are still obtainable with moderate to high degrees of certainty (e.g. sedan, white color, and make/model based on contours and grill pattern). Though, SD is clearly less detailed, it can be argued that in practical terms they deliver roughly the same probability of identifying vehicle make/model and color.

Midday Intersection (Far Human)

In this scenario, cameras are digitally zoomed into a human subject ~135ft out, corresponding to a ~126ft HFoV. Midday lighting conditions are even and not particularly challenging.

At this fairly wide FoV, the higher MP (namely 5MP and 2MP) and 720p cameras provide key details that SD is unable to deliver. Although, identification is considerably unlikely, other key details such as complexion, gender, clothing, build, and shaved head are obtained with fair degrees of certainty. In contrast the SD cameras render the human subject as an essentially non-descript 'blob'.

Midday Intersection (Near Human)

In this scenario, cameras are digitally zoomed into a human subject ~35ft out, corresponding to a ~33ft HFoV. Midday lighting conditions are even and not particularly challenging.

In contrast to the 'far' scenario (human), all cameras including SD, provide a moderate to strong sense of complexion, gender, clothing, build, and shaved head. Several of the MP cameras, especially the 2MP, elucidate enough details to make identification a moderate to strong likelihood as well.

Nighttime Intersection (Far Vehicle)

In this scenario cameras are digitally zoomed into two vehicles located ~130ft out, corresponding to a HFoV of ~121ft. The nighttime environment produces some lighting challenges due to head-lights directed into the camera rig. In general the scene is illuminated by transient vehicles and static street lighting (~5 to ~15 lux).

The head-lights from the vehicles are posing a considerable problem for the 5MP D/N (Axis P1347) and 1.3MP D/N (Arecont AV3135). Due to blooming head-lights, these cameras produce an image that is severely washing out the vehicle(s) of interest. It is also clear that none of the cameras provide an ability for plate capture or other identifying details that would lead to the make/model. Interestingly, the two cameras in color mode (2MP IQ042SI and SD Axis M1011) provide among the best picture quality.

Nighttime Intersection (Mid Vehicle)

In this scenario cameras are digitally zoomed into a vehicle located ~85ft out, corresponding to a HFoV of ~79ft. The nighttime environment produces some lighting challenges due to transient vehicle lights. Note also that the vehicle is located in an area where light reflecting off the moist street surface causes additional lighting challenges. In general the scene is illuminated by transient vehicles and static street lights (~5 to ~15 lux).

It is quite evident that none of the cameras provide enough detail for plate capture. The 5MP D/N (P1347) and 1.3MP D/N (AV3135) images appear considerably degraded or washed out - due not only to blooming head-lights but also from, perhaps, other light sources (e.g., reflected light from surface moisture, and the vehicle's reflective surface itself). Similar to the previous 'far' scenario, the color mode cameras (2MP IQ042SI and SD M1011) provide somewhat better than expected results. Also, interestingly, the other SD (Bosch NBN-498) is delivering crisper vehicle contours than many of the other cameras in the composite.

Nighttime Intersection (Near Vehicle)

In this scenario cameras are digitally zoomed into a vehicle located ~50ft out, corresponding to a HFoV of ~47ft. The nighttime environment produces some lighting challenges due to transient vehicle lights and reflective surface moisture. In general the scene is illuminated by transient vehicles and static street lights (~5 to ~15 lux).

In contrast to its daytime counterpart, this nighttime composite exposes much less vehicle details. While the contours of the vehicle are fairly crisp in several of the cameras (2MP, 720p NBN-921, SD M1011, and SD NBN-498), the grill pattern of the vehicle is quite blurry across the entire composite. Identifying make and/or model is considerably unlikely, given the lack of clarity in these images. Note, again, that the 5MP D/N and 1.3MP D/N experienced the most image quality degradation due to blooming light-sources.

Nighttime Intersection (Far Human)

In this scenario cameras are digitally zoomed into a human subject located ~135ft out, corresponding to a HFoV of ~126ft. The nighttime environment produces some lighting challenges due to transient vehicle lights and reflective surface moisture. In general the scene is illuminated by transient vehicles and static street lights (~5 to ~15 lux).

In contrast to its daytime counterpart (Midday Intersection 'Far' Human), this nighttime composite provides much less detail. For instance details such as complexion, clothing, gender, and hair-style are not readily discernible. However, gross contours on the subject, arguably reveal a taller/medium build. In terms of these contours, the B/W mode cameras provide greater clarity and contrast versus the color mode cameras. Note especially that in the SD color-only (Axis M1011) the subject is essentially a 'blob' and blends in with the surroundings due to a lack of contrast.

Nighttime Intersection (Near Human)

In this scenario cameras are digitally zoomed into a human subject located ~35ft out, corresponding to a HFoV of ~33ft. The nighttime environment produces some lighting challenges due to transient vehicle lights and reflective surface moisture. In general the scene is illuminated by transient vehicles and static street lights (~5 to ~15 lux).

The color mode cameras (2MP IQ042SI and SD Axis M1011) provide clear advantages over the b/w mode cameras in this composite. Most notably, a stronger indication of the subject's color of clothing and complexion is provided by the color mode cameras. Interestingly, the color mode cameras now provide more contrast than the b/w cameras since the subject 'moved' from 'far' to 'near'. In practical terms, among the D/Ns in b/w mode there is little differentiation. They tend to all provide fair indications of gender, build and hair-style. Identification of the human subject is highly unlikely across all cameras in this composite, as well.

Narrow Half Intersection

Covering a complete 6 x 6 lane intersection is clearly challenging even for megapixel cameras. As a final experimental test, we tried covering only half the intersection to see what improvements in details would be captured. As the capture below shows, a fairly dramatic difference in details appear. Facial details can be made of subjects at the corner close to the camera where previously only a silhouette appeared. Even across the intersection, personal details can be captured as well as an outline of license plates. The big problem here is how realistic only covering half an intersection is. Most users will want to deploy a single camera to cover the whole intersection.

Methodology

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

  • Arecont AV3135 (online $883) - 3MP/1.3MP D/N; 1/2" CMOS; Tamron F1.6; 0.1 Lux (Color)
  • 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 (online $775) - 720p D/N; 1/3" CCD; F1.3 Bosch; 0.4 Lux (BW)
  • Bosch NBN-498 (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 two (2) key lighting scenarios. All cameras are set to their defaults during test scenarios except at night when the D/N cameras were forced to night 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. During the nighttime scenario D/N cameras were forced into b/w mode.

Here are the key default settings for each camera:

  • 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 (2) key lighting scenarios we examined:

  • Midday Even Lighting - Even sunlight ~4,000 lux
  • Nighttime - Artifical lighting varies from ~5 lux to ~10 lux

For each lighting scenario, a human subject walks from the far corner of the scene located ~155ft from the camera rig. The human subject first walks right to left through a 'horizontal' crosswalk and then another crosswalk from back to front. The subject finishes at the end of the 'vertical' crosswalk (~30ft). Vehicle subjects are chosen from the natural traffic that occurs during the filming of the human subject's transit from 'far' to 'near'.

Methodology (Half-Intersection)

The Half-Intersection test looks to discover trade-offs when decreasing the HFoV ~1/2 to monitor an intersection. The following camera is used for the Half-Intersection test:

  • Axis Q1755 (online $1400) - 720p; 1/3" CMOS; lens F1.8 - F5.1; 0.2 Lux (BW); 2.0 Lux (Color)

This test focuses on one (1) key lighting scenario:

  • Midday Even Lighting - Even sunlight ~4,000 lux

For this lighting scenario, a human subject walks from the far corner of the scene located ~155ft from the camera rig. The human subject first walks right to left through a 'horizontal' crosswalk and then another crosswalk from back to front. The subject finishes at the end of the 'vertical' crosswalk (~30ft). Vehicle subjects are chosen from the natural traffic that occurs during the filming of the human subject's transit from 'far' to 'near'.

1 report cite this report:

Directory of Camera Shootout Series on Dec 27, 2011
The following directory lists all of our camera shootouts. These shootouts pit 4 - 8 different surveillance cameras in simultaneous tests on real...
Comments : PRO Members only. Login. or Join.

Most Recent Industry Reports

July 2018 IP Networking Course on Jun 22, 2018
  This is the only networking course designed specifically for video surveillance professionals. Register now. Lots of network training exists...
Installation Hardware for Video Surveillance - Indoor Fasteners on Jun 22, 2018
As part of our Installation for Video Surveillance series, in this note, we cover drywall anchors. A key part of installing security hardware is...
Hikvision ColorVu Integrated Visible Light Cameras Examined on Jun 22, 2018
When it comes to low light, infrared light has become the defacto standard in surveillance. But IR is limited to monochrome images, making colors...
'Secure Channel' OSDP Access Control Examined on Jun 21, 2018
Despite claiming to be better than Wiegand, OSDP's initial releases did not address the lack of encryption between reader and controller, leaving...
Most Wanted Improvements In Manufacturer Technical Support (Statistics) on Jun 21, 2018
5 key areas of improvement and 1 clear wanted support feature were voiced by 140+ integrator responses to: What improvement in manufacturer...
GDPR / ICO Complaint Filed Against IFSEC Show Facial Recognition on Jun 20, 2018
IPVM has filed a complaint against IFSEC’s parent company UBM based on our concern that the conference violates core GDPR principles on...
IFSEC 2018 Final Show Report on Jun 20, 2018
IPVM attended the IFSEC show for the first time this year. The Chinese took over the show, centered on Hikvision, flanked by Dahua, Huawei and a...
Mobotix Releases 'Move' Into 21st Century on Jun 20, 2018
For years, Mobotix stood resolutely against, well, every other manufacturer, selling it as a virtue: MOBOTIX equipment is designed with no...
Cybersecurity Startup VDOO Disclosing 10 Manufacturer Vulnerabilities Starting With Axis And Foscam on Jun 20, 2018
Cybersecurity startup VDOO has uncovered significant vulnerabilities in Axis cameras along with many others not yet disclosed. In this report, we...
Axis Guardian - Cloud VMS And Alarm Monitoring - Released on Jun 19, 2018
Axis has struggled to deliver a cloud-based managed service video platform. Video service providers have utilized AVHS for over a decade, and have...

The world's leading video surveillance information source, IPVM provides the best reporting, testing and training for 10,000+ members globally. Dedicated to independent and objective information, we uniquely refuse any and all advertisements, sponsorship and consulting from manufacturers.

About | FAQ | Contact