Miovision worked out some bugs with uploading video to them so I can start my testing of their traffic counting system. My initial tests are promising, but I have A LOT more testing to do before I am sure their system is good to go in every situation.
Test 1 – Rainy day with wind gusts up to 35 mph:
Intersection 1 is a large suburban intersection (four lane divided highway, dual lefts on one approach, turn lanes all around, free right islands on all four approaches). Intersection 2 is a smaller intersection with turn lanes, but each approach just has one through lane. There are no driveways or intersections in between the two intersections. RESULTS – over the two hour am period there is a 1% difference in northbound volumes between the intersections and a 4% difference in the southbound volumes between the intersections. Rain drops stuck to the lens in the p.m. and Miovision couldn’t accurately process the video. They tell me they are working on a new camera housing to minimize this issue.
Test 2 – Good weather:
Intersection 1 is a single lane roundabout, single lane approaches all around. Intersection 2 is a tee intersection with the mainline having one turn lane and one through lane on each approach. The minor leg has one approach lane There are no driveways or intersections in between the two
intersections. RESULTS – over the two hour pm period there is a 1%
difference in eastbound volumes between the intersections and a 1%
difference in the westbound volumes between the intersections. We didn’t collect am data.
So far, these are very encouraging results. I am going to do some manual ground truthing on test videos in the future. I am interested in opinions on how to slice the data to prove accuracy. Due to slight variations in clocks, travel distance between intersections, etc. there are bigger discrepancies at the 15 minute and smaller intervals. I would consider a 95% accurate peak hour to be good enough. Thoughts?