Testing Panoramic Cameras Outdoors

Outdoors, PTZs tend to be a favorite for covering large open areas but they only see whatever they are currently pointed at. Panoramics are an emerging alternative but how well do they perform and what type of video quality do they deliver?

Following our groundbreaking panoramic camera shootout, we took a group of panoramic cameras to a moderately busy intersection scene to test their performance viewing pedestrians and vehicle traffic at various distances, shown in this overhead view:

The cameras were mounted on our portable tripod, about 13' off the ground:

Delivering panoramic overviews like so:

Here are our key findings from the test:

• Even in a well lit daytime scene, recognition of a human subject is possible only at very short range, e.g., a person waiting to cross the street. As we found in our indoor shootout, distances further than 5-10 feet provide little in the way of identifiable details.
• Even at the shortest range possible, with a car just past the cameras, license plates were not readable. However, make, model, and color of vehicles is easily captured.
• Performance varied widely at night. The Axis M3007 captured our human subject better at longer distances, but suffered from dropped frames and handled vehicle lights poorly. Panasonic handled lights best, but the human subject worse due to a dim image. Finally, the Vivotek FE8172V produced the brightest image, but noise and worse light handling made our subject difficult to spot and vehicles difficult to recognize.
• At night, recognition even at short distances is difficult, due to noise and dimmer images. Detection of a subject at medium to long ranges (50'+) also becomes difficult. Vehicle information becomes very difficult to gather, due poor handling of headlights, noise, and dim images. Make and model were not possible to determine in most cases, and color was practically impossible to determine.

Taking the above into account, we believe the best use for panoramic cameras outdoors is situational awareness only. Users may find them desirable in conjunction with other cameras, to provide an overview of happenings in the scene, and provide the ability to follow subjects or vehicles as they move throughout the scene. We do not recommend using panoramic cameras if any level of identification is a requirement, aside from vehicle make and model in well lit scenes.

Daytime Comparisons

In order to test foot traffic common at intersections, we had our subject walk from approximately 100' to near the crosswalk signal, about 8' from the cameras. Starting with this overview image, we can immediately see differences in color and brightness among the three cameras. Some of this may be accounted for by switching white balance mode, but all were left on automatic for out tests.

The other notable difference is how much of the lens each camera uses. We can see the fuzzy edge of the fisheye in the M3007, while Panasonic crops this out. Vivotek does not crop it, but does not let the user pan up far enough to see it. This can have an effect on image quality, as resolution worsens the further you move towards the edge of the image, evident in the blurring in the Axis image. However, eliminating too much around the fisheye reduces the overall coverage of the camera. The FE8172V is an example of this, as this image is panned as far up as possible, but the image cuts off before reaching the horizon, visible in the other two images, despite all cameras being mounted at the same height and leveled.

At 100', the subject is visible, without any sort of fine details. Light shirt and (at best) blue jeans is all that can be made out.

Moving to about 50', at the crosswalk before crossing the street, slightly more detail is available on some cameras, though not much. At this point, we can also note differences in dewarping, with the subject appearing squat on the Axis M3007, and approximately proportionate on the other cameras. This could be an issue if trying to gauge height of a subject.

At 25', we can begin to make out facial features. At this point the subject may be recognizable if he's known, but strangers would be hard to identify with this resolution. The subject still appears wider using the Axis camera than others.

Finally, at 8', the subject is recognizable, with glasses and hair visible. A logo is almost visible on his shirt, as well.

Make and model of vehicles moving through the intersection were not difficult to determine, but even at the closest possible range (~20'), license plates were not discernable:

This clip shows traffic in motion for comparison. Users may download this video clip, as well. Note that both the M3007 and WV-SF458 dewarped on the camera, so the view may not be changed. The FE8172V features dewarping and immersive controls, however, so users may view the whole scene using this archive player download.

Nighttime Comparisons

Performance in the same test at night varied. The lighting in this scene varied between approximately 5-30 lux, depending on how close our target was to street and traffic lights, and discounting the effects of headlights. All cameras generally provided enough quality to monitor activities, though details of either our human subject or vehicles were difficult to detect. The Axis M3007 produced the clearest image with the truest color and least noise. The Panasonic WV-SF458 handled headlights and taillights best, but was overall the dimmest of our cameras, with the subject difficult to detect at range. Finally, though the FE8172V is the only true day/night camera in this test, and it provided bright images, it handled headlights and taillights poorly, and had substantial noise compared to the other two.

At 100', our subject is barely detectable, at best:

At 50', the subject is discernable on the M3007, though without any detail. He remains difficult to spot in the other two cameras, do to light levels and/or noise:

At 25', clothing can be at least partially determined using the M3007, and he is easily detectable.

Finally, at close range, clothing becomes easy to determine, and some facial detail is available.

Car color aside from "light" or "dark" was difficult to determine due to glare from headlights and taillights, making color a moot point. Make and model were practically impossible to determine, as well, due to noise, glare, or motion blur.

This video shows these effects in the nighttime traffic scene. Users may also download a 20-second video clip.

Test Details

We tested three cameras from our panoramic shootout:

• Axis M3007-PV -- 5MP sensor, cropped to 800x600 view area, dewarped on camera
• Panasonic WV-SF458: 3.1MP sensor, cropped to 1.3MP view area, dewarped on camera
• Vivotek FE8172V: 5MP sensor, dewarped client side

All settings were left defaulted, except shutter speed, which was standardized at 1/30s. Recordings were at the camera's highest possible frame rate and resolution.