I fucking ABHOR protected left turn arrows

[tradephoric]

Stuck-on calls can be problematic for a few reasons. 

1.  A stuck-on call can unnecessarily cause a phase to run when a vehicle is no longer on the detection zone. 

2.  For adaptive signals, a stuck-on call can lead to inaccurate split-voting.  A minor side-street may receive the lions share of the green time and squeeze the main phase if a side-street detector is sticking on (the controller thinks a vehicle is sitting on the side-street the entire phase indicating very heavy traffic, when in fact no vehicle is there). 

[tradephoric]

The traffic cameras sense a change in contrast in a particularly defined zone on the screen, so it's not just being able to "see" a vehicle or better ambient lighting making detection better. Even if the car is black and sitting on black pavement, the headlights at night should be enough to trip the detection.

You are basing your opinions on logic.  Unfortunately, the performance of video detection can be illogical.  Sure, in theory video detection should work great... it doesn't though.

[jakeroot]

You are basing your opinions on logic.  Unfortunately, the performance of video detection can be illogical.  Sure, in theory video detection should work great... it doesn't though.

I think you're basing your opinions on too few signals. As I said before, most of the signals near where I live are operated via video detection, and I have not noticed any major issues (other than once many moons ago, and even then, the issue was that it just didn't notice me).

On the other hand, cycling is now a big deal and I'm not entirely convinced in-ground plates are capable of detecting cyclists. In places like Seattle, cycling is a major form of transportation, and they now have to run red lights just to get through signalized intersections. Obviously, ignoring signals is undesirable (to say the least). Can video detection detect cyclists? I don't know. But I think the chances are higher for them than the in-ground plates.

[tradephoric]

I am basing my opinions on facts.  Here is a summary of just some of the research done analyzing the accuracy of video detection.  The overwhelming theme is that video detection is not as accurate as inductive loops.  That's a fact. 

A report from MacCarley (1998) compared the performance of the Vantage Video Traffic Detection System in a wide variety of weather and illumination conditions.  The report concluded that under ideal or near ideal traffic, camera configuration, and illumination conditions, the ability to properly detect a vehicle was adequate. It noted, however, that even under optimal conditions false calls were common. The average presence detection accuracy over all data sets was 80.9% with false detection errors of 8.3%.

The Illinois Center for Transportation (2009) tested 3 different VDS systems under a wide variety of illumination conditions. During sunny morning conditions, false calls increased for all 3 systems when compared to the base "cloudy noon"  condition: Autoscope increased from 3.5% to 11.2%, Peek from 7% to 14.4%, and Iteris from 5.7% to 20.7%. Additional trends were seen during other lighting conditions with false calls increasing during dawn for all three VDS (up to 23%), and increased false calls were observed at night due to the reflection of headlights on the pavement.

A report by the Federal Highway Administration (2006) analyzed Iteris video detection along a 21-mile stretch of Virginia State Route 7. The report found that that VDS consistently over-counts traffic volumes, consistently missed left-turn phase demands, had frequent occlusion issues along side-street detection, and experienced significant data errors due to glare in sunrise and sunset conditions. As part of the recommendations, it was suggested that the cameras along the route be repositioned to shoot from the near-side of the intersection (looking nearly straight down onto the detection zone) as opposed to shooting across from the far-side of the intersection.

A technical study by Rhodes (2005) found that VDS-based detectors statistically produce a higher number of false detections and a higher number of missed detections than inductive loops. The report cites that lighting is the most common condition for causing video detection errors, especially during transitional periods at sunrise and sunset. The effects of vehicle occlusion can lead to either missed detections (vehicle blocking another vehicle from being detected) or false detections (vehicle placing a call on adjacent lane when no vehicle present).

[riiga]

Can video detection detect cyclists? I don't know. But I think the chances are higher for them than the in-ground plates.

Why not just use induction loops? Or is that the same as in-ground plates? They're used at pretty much every intersection I've encountered here, and work for bikes too.

[KEK Inc.]

Can video detection detect cyclists? I don't know. But I think the chances are higher for them than the in-ground plates.

Why not just use induction loops? Or is that the same as in-ground plates? They're used at pretty much every intersection I've encountered here, and work for bikes too.

Standard induction loops often don't detect motorcycles or bicyclists.  Some municipalities install a more sensitive loops on a marked spot.  Making the roadway induction loops more sensitive could post problems for false-positives triggering the light.

[tradephoric]

The inaccuracy of video detection is well documented.  If a 40-ton semi doesn't get detected by video detection what makes you think a 20 lbs bicycle will?  Admittedly, inductive loops can have difficulty detecting bicyclists but so can video detection.

EDIT:  Just to be clear, it's not the weight of the vehicles that matter in the above sentence.  If a big honking semi doesn't get detected by the camera then it's not looking good for the bicycle which has a much smaller footprint.

[tradephoric]

I've never had any issue at video detection signals anywhere in Nevada or California.

Have you ever been to a video detection signal anywhere in Nevada or California? 

[Scott5114]

The City of Norman uses video detection almost exclusively, and I rarely have problems with them. I did have a recently-installed signal appear to fail to detect me recently, but I rarely pass through that intersection and I'm not sure if was actually a failure with the signal or if there's wonky timing programming to blame.

The intersection I have the most problems with is an ODOT-administered signal that uses induction loops. I think the loops were damaged or obscured by recent repaving. Either that or they have some horrendous lag time programmed in. I usually have to sit at the intersection and creep forward a few times before the signal cycles, usually at least a full minute after I've been sitting there.

One thing that I've had success with at video detection signals is flashing my brights for a moment. If the signal is not detecting me, the sudden spike in brightness is often enough to get its attention.

[roadfro]

I've never had any issue at video detection signals anywhere in Nevada or California.

Have you ever been to a video detection signal anywhere in Nevada or California? 

I'm from Las Vegas and live in Reno...so yeah, I've been through more than a couple video detection signals in Nevada... Video detection has been widely utilized in new signal construction throughout Nevada for the last decade or more.

The rate of new signals in Northern Nevada utilizing video detection has dropped in recent years, but there's still plenty around. I can't say with certainty whether this decline has been due to performance issues (NDOT was doing a study a few years ago out of UNR analyzing effectiveness of various video vendors equipment in multiple light/weather conditions) or to reconstruction (virtually all new or reconstructed signals in Reno/Sparks in recent years have accompanied pavement [re-]construction, when it's easy to put in loops).

However, all the jurisdictions in the Vegas area are still very much on board with video detection. I can only think of one new signal in the past 5-ish years that isn't using video (Las Vegas Blvd at Harmon Ave, where the intersection is so wide and skewed that typical camera positioning would likely be ineffective). Most newer installations have the detection camera mounted along the mast arm on a riser pole (between the left turn and through signal heads), which I assume is done to help alleviate some of the occlusion issues. Most cameras, particularly those facing east/west, also usually have a longer visor to help reduce issues with sun glare during sunrise/sunset.

[tradephoric]

I think you're basing your opinions on too few signals. As I said before, most of the signals near where I live are operated via video detection, and I have not noticed any major issues (other than once many moons ago, and even then, the issue was that it just didn't notice me).

Aren't you basing your opinions on too few signals?  For one, faulty detection can be easily masked by tying down the signal.  Many signals have minimum recalls and will cycle a phase regardless of detection.  In addition, a vehicle that is missed by video detection may still get serviced due to the detection of surrounding vehicles (a vehicle waiting on EB side-street who is getting missed may still get a green because the vehicle on WB side-street is being detected).  You can't base your opinions on the accuracy of video detection simply because you "didn't notice any major issues"  while driving through the intersection.  I'm citing statistically significant studies detailing the performance issues with video detection .  I'm the last person that should be accused of providing anecdotal evidence and basing my opinions on "too few signals"  in this thread. 

However, all the jurisdictions in the Vegas area are still very much on board with video detection. I can only think of one new signal in the past 5-ish years that isn't using video (Las Vegas Blvd at Harmon Ave, where the intersection is so wide and skewed that typical camera positioning would likely be ineffective). Most newer installations have the detection camera mounted along the mast arm on a riser pole (between the left turn and through signal heads), which I assume is done to help alleviate some of the occlusion issues. Most cameras, particularly those facing east/west, also usually have a longer visor to help reduce issues with sun glare during sunrise/sunset. 

A report by the Federal Highway Administration (2006) analyzed Iteris video detection along a 21-mile stretch of Virginia State Route 7. The report found that that VDS consistently over-counts traffic volumes, consistently missed left-turn phase demands, had frequent occlusion issues along side-street detection, and experienced significant data errors due to glare in sunrise and sunset conditions. As part of the recommendations, it was suggested that the cameras along the route be repositioned to shoot from the near-side of the intersection (looking nearly straight down onto the detection zone) as opposed to shooting across from the far-side of the intersection.

It's important to note that some agencies do shoot from the near-side of the intersection, with the camera looking nearly straight down onto the detection zone.  Sun glare and inaccurate detection during sunrise/sunset conditions is not rectified by shooting straight down.  Shooting straight down works great in a cloudy noon conditions, but fail under sunrise/sunset conditions (just like cameras that shoot across the intersection that are pointed more towards the horizon).   I have a feeling that simply extending the visor isn't going to solve the sun glare issue either.

The rate of new signals in Northern Nevada utilizing video detection has dropped in recent years, but there's still plenty around. I can't say with certainty whether this decline has been due to performance issues (NDOT was doing a study a few years ago out of UNR analyzing effectiveness of various video vendors equipment in multiple light/weather conditions) or to reconstruction (virtually all new or reconstructed signals in Reno/Sparks in recent years have accompanied pavement [re-]construction, when it's easy to put in loops).

I'd be interested in the results of UNR's study.