Vehicle trigger settings - missing subsequent vehicles after initial capture

[Nidstang]

My LPR camera detects single vehicles well. But when vehicles pass in quick succession, while BI will trigger consistently for each vehicle, the footage sent to CodeProject.AI has missed vehicle #2, #3, #4 etc. They have already passed and are out of frame.

The second vehicle is triggering anywhere from 2 to 5 seconds after the first vehicle, but is absent from the analysed footage. Whereas a fresh trigger with nothing preceding it always captures the vehicle nicely.

One of the AI data files for a following vehicle is seen below. It saw the vehicle at T-1.5s when BI triggered, but the analyses starts at T+0s i.e. 1.5s later when the vehicle is gone.

I have a 3s pre-trigger record time, because when viewing the clips I like to see the vehicle come into frame, but I'm unsure if this affects things or how BI and CodeProject handle overlapping 'alert' clips. My settings are below, appreciate any input. Thanks.







Note: I just noticed I have 'continuous sub + Triggered' set but I am only using Main Stream on this camera. Not sure if that matters.

 

[wittaj]

Make your end trigger number much shorter. I use 0.1 seconds.

 

[Nidstang]

Are you referring to Break Time > End trigger unless re-triggered within? The minimum I can set mine to is 1.0 sec, it won't go any lower
I'm running Blue Iris 5.8.2.5 if that matters.

 

[actran]

@Nidstang Enable "Begin analysis with motion-leading image" to see if that helps change the first image that is processed.

Also, from your screenshot, it's taking CP.AI ~900+ msec to process each image. Disable "Use main stream if available" will reduce CP.AI processing time so it can process more images faster. Based on your camera AI configuration, I assume you just want to capture the plate image, and not actually read/extract the plate #?

 

[wittaj]

Are you referring to Break Time > End trigger unless re-triggered within? The minimum I can set mine to is 1.0 sec, it won't go any lower
I'm running Blue Iris 5.8.2.5 if that matters.

Bummer - I am running an older version and have make time as 0sec and the Break Time>End Trigger at 0.1seconds

 

[Nidstang]

@Nidstang Enable "Begin analysis with motion-leading image" to see if that helps change the first image that is processed.

Also, from your screenshot, it's taking CP.AI ~900+ msec to process each image. Disable "Use main stream if available" will reduce CP.AI processing time so it can process more images faster. Based on your camera AI configuration, I assume you just want to capture the plate image, and not actually read/extract the plate #?

Thanks I'll try enabling that one. For my LPR camera I don't have sub-stream enabled. My thinking is if it fails to trigger on a vehicle, I can still go back and manually read the plate. I do have CP.AI reading/extracting the plate number and saving it as text on the alert. I could enable sub-stream, and still record main stream continuously (and uncheck 'use main stream if available') - but would this mean CP.AI will only analyse the sub-stream resulting in lower accuracy?

Does the longer processing time mean it's analysing subsequent footage too late? I assumed BI sends each alert as it happens and they just queue up in CP.AI until they are processed.

I have to admit I know nothing about the text parameters entered in the fields below the 'CodeProject.AI' tickbox. I simply copied a configuration from the forums here but it seemed to work lol.

Bummer - I am running an older version and have make time as 0sec and the Break Time>End Trigger at 0.1seconds

Weird, I wonder why they changed it? Can I ask what version you are running?

 

[wittaj]

Keep in mind that CPAI downrezes your mainstream image before processing, so you are simply saving it a step by feeding it the substream image to begin with and that does speed up the results a bit.

If the CPAI processing times are too high, it will just timeout and flush out and that could be why some are missed, but I think the main reason is because it looks at each trigger as ONE event, so if two vehicles pass during that break time, it will only find one, thus having short break times is key so that the 2nd vehicle is a new event.

I am on 5.5.6.21, so there are a lot of versions where that could have been changed.

 

[Nidstang]

@wittaj The more I think about it, the more I think that break time setting is to blame...

On a slight tangent, my LPR camera is at an angle of about 35 degrees to the road. I assume this makes it easy to trigger because vehicles pass from in and out of the screen fairly quickly. I'm planning to move the LPR camera almost straight on with the road, say 5 degrees. Will triggering be difficult like this? The vehicles will be coming towards/away from the camera, so growing/shrinking in size as opposed to moving across the image. I feel like it would be difficult to trigger at an optimum distance, leading to CP.AI having to process much more footage.

I don't need to solve this, but there is a bit of work to relocate the camera with the assumption it will perform better in the new spot, so would be good to know if I'm completely wrong about that!

 

[Nidstang]

Keep in mind that CPAI downrezes your mainstream image before processing, so you are simply saving it a step by feeding it the substream image to begin with and that does speed up the results a bit.

If the CPAI processing times are too high, it will just timeout and flush out and that could be why some are missed, but I think the main reason is because it looks at each trigger as ONE event, so if two vehicles pass during that break time, it will only find one, thus having short break times is key so that the 2nd vehicle is a new event.

I am on 5.5.6.21, so there are a lot of versions where that could have been changed.

Thanks, I'll see if I can reduce the processing times. Do we know what resolution CP.AI works with so we can match up? If it works like that...

 

[wittaj]

I believe it drops it to 640x480 resolution.

Regarding the camera location and triggering, that is where it is hard to find the optimal spot.

I have one that is a big angle like yours and because of the angle and make time, it is RARE that two plates would be in the field of view during the same trigger.

A more straight on shot gives you a better chance to capture a clean plate, but the plate is also in the field of view longer, so there are chances that multiple plates could come thru on the same trigger and only one is captured.

Then it is a balance of using zones to try to force break times and have each plate be its own trigger, but get too cute with the zones and then you may miss a trigger also.

 

[actran]

@Nidstang Are you getting plate #s with your AI configuration (in last shared screenshot)? (hence my surprise)

If you are not getting extracted plate #s, then you should try MikeLud1 config: Blue Iris and CodeProject.AI ALPR

 

[Nidstang]

A more straight on shot gives you a better chance to capture a clean plate, but the plate is also in the field of view longer, so there are chances that multiple plates could come thru on the same trigger and only one is captured.

Then it is a balance of using zones to try to force break times and have each plate be its own trigger, but get too cute with the zones and then you may miss a trigger also.

Thanks for the feedback. I have a Plan C location that is much closer to the road but still at a bit of an angle, so I will consider that. I'll also have an overview cam so might be able to use that to trigger the LPR.

@Nidstang Are you getting plate #s with your AI configuration (in last shared screenshot)? (hence my surprise)

If you are not getting extracted plate #s, then you should try MikeLud1 config: Blue Iris and CodeProject.AI ALPR

Yep it's working as is It's extracting plate numbers and burning them onto image captures. I found those settings in the same thread but earlier on. I'll try Mike's updated settings with the asterisks. It sounds like it won't help with me not capturing successive vehicles, that seems to be a result of my BI settings.