I checked the explorer and have been doing several QA reviews of these images and a vast majority of the images I'm providing quality assurance on fail QA my visual test. About 85% fail to the 15% that have some decent quality.
I recommend hivemapper implement an automated QA test. Rather than developing your own, consider using http://open.qa/ This QA system can be easily solve solve some verification issues by using the needles function. This will help the project scale images quickly to https://hivemapper.com/explorer and likely would solve bug #82 Jenkins could possibly be another solution, but I know that openQA is much more build for your propose. You can make needles that have a percentage aspect that is acceptable. This would automatically detect hoods, lights and other items inside the windshield that can specifically be targeted.
I understand you are building a live view in the app to see your imagery, which is great because it will solve a lot of issues up front. The percentage of pics that have passed through QA would be good, but is probably a long-term project. I would recommend that at least the wallet allows for users to see the images they take so they can at least make quality adjustments to the device so the margins of error begin to close the gap.
I checked the explorer and have been doing several QA reviews of these images and a vast majority of the images I'm providing quality assurance on fail QA my visual test. About 85% fail to the 15% that have some decent quality.
I recommend hivemapper implement an automated QA test. Rather than developing your own, consider using http://open.qa/ This QA system can be easily solve solve some verification issues by using the needles function. This will help the project scale images quickly to https://hivemapper.com/explorer and likely would solve bug #82 Jenkins could possibly be another solution, but I know that openQA is much more build for your propose. You can make needles that have a percentage aspect that is acceptable. This would automatically detect hoods, lights and other items inside the windshield that can specifically be targeted.
I understand you are building a live view in the app to see your imagery, which is great because it will solve a lot of issues up front. The percentage of pics that have passed through QA would be good, but is probably a long-term project. I would recommend that at least the wallet allows for users to see the images they take so they can at least make quality adjustments to the device so the margins of error begin to close the gap.