I ordered one of the Motorola cameras that @TheWaterbug mentioned previously. Hoping it might contain some free billion dollar magic for 30 dollars...
What I have noticed with the lyorta's is that they are very hesitant to switch IR/daylight modes. You really need to set a manual schedule in Blue Iris. I missed a whole lot of plates between 5-8am because of this and never really addressed it until recently.
If Motorolas are consistently available, that's gotta be a killer solution, but otherwise (or for any lyorta/dahua users [most]), maybe an addition or revision to the cliff notes could be helpful. I know there are threads talking about settings, but I myself have combed through them without much clarity in the past. A more detailed dummy's guide to ALPR could be nice for a lot of folks.
Edit:
Also, if there is, in fact, some sort of special image processing embedded in the Motorola devices, coupling that with the power of crowdsourced global residential image collection could create a near professional-grade LPR solution... The fact that these devices are available secondhand and that we are able to collect data that they otherwise couldn't really is an exciting prospect.
I've seen both IRL and on video, so many examples of lighting-fast inference and seemingly excellent accuracy in the police car systems. It seems like we aren't that far away if we can get a large enough training set. While @MikeLud1 understands the inner-workings of the model far better than I do, we have discussed a fair amount, and it seems like we have plenty of resources to train a model on a very very large data set once we collect enough images. My hope is that we can at least get close to what the commercial solutions are offering.
What I have noticed with the lyorta's is that they are very hesitant to switch IR/daylight modes. You really need to set a manual schedule in Blue Iris. I missed a whole lot of plates between 5-8am because of this and never really addressed it until recently.
If Motorolas are consistently available, that's gotta be a killer solution, but otherwise (or for any lyorta/dahua users [most]), maybe an addition or revision to the cliff notes could be helpful. I know there are threads talking about settings, but I myself have combed through them without much clarity in the past. A more detailed dummy's guide to ALPR could be nice for a lot of folks.
Edit:
Also, if there is, in fact, some sort of special image processing embedded in the Motorola devices, coupling that with the power of crowdsourced global residential image collection could create a near professional-grade LPR solution... The fact that these devices are available secondhand and that we are able to collect data that they otherwise couldn't really is an exciting prospect.
I've seen both IRL and on video, so many examples of lighting-fast inference and seemingly excellent accuracy in the police car systems. It seems like we aren't that far away if we can get a large enough training set. While @MikeLud1 understands the inner-workings of the model far better than I do, we have discussed a fair amount, and it seems like we have plenty of resources to train a model on a very very large data set once we collect enough images. My hope is that we can at least get close to what the commercial solutions are offering.
Last edited:
