Hi,
Anyone know any solution for people counting vehicle occupants during both day and night?
Thanks
Hi,
Anyone know any solution for people counting vehicle occupants during both day and night?
Thanks
Probably the cheapest option, unless the clicky-counter interface belongs to a union.
I had a design project for a major University quite a few years back that had this exact requirement. They had a parking lot designed exclusively for students who carpooled, and wanted to be sure that every car passing through the parking gates had at least two occupants. Students soon discovered that they could register for the program to gain access to the lot, but then use it even when they were commuting alone.
The two gates to enter the parking lot were card reader controlled, so we first considered requiring two consecutive card swipes from different students cards to open the gate. This was thought to have the potential to create traffic backups, so was immediately rejected by the client.
The second plan was to put cameras at the gate, one to look into the car, and a second to capture the license plate. The video would be recorded, and then staff members would later review the recordings in fast forward in an attempt to catch violators. Most cars entered the parking lot during a three-hour time window, so the auditing of the recorded video was a manageable task.
The client liked the idea, and we ended up doing a full set of plans and specifications for the project, however the system was never installed. I think that they eventually ended up stationing an employee at the gates to periodically monitor incoming traffic.
At the time, video analytics was so immature that I never even considered using it for this application. I would be surprised if it was that much better today.
Camio was said to have very good ability to recognize and label humans within a scene. Not sure if it could be successful with only head and shoulders... and it didn't appear to have LPR... still...
I wonder if its human labelling could work in conjunction with scripted metadata summation to support this need?
Mr. Maslan, could this be a bridge too far?
Camio labels that fact that video events have people in them, but doesn't yet count the number of individual people in the scene.
If you could see eyes/nose/mouth, then counting can be pretty accurate, but otherwise the problems with windshield reflections/tint/etc... can make counting nearly impossible unless you have a great camera view of passengers. You can see from this Hikvision test camera facing the road below our office that it's really hard even for human eyes to count passengers as cars pass: https://www.camio.com/c/81o6oom6c4jk
Bottom line: if the camera has the perspective of a close-up gate guard, and the challenge were to answer with 96% precision "is there more than one person in the car?", then we'd accept that challenge. Do you have any sample video you could share?
Camio has OCR, so the text recognition depends on the clarity of the image. For example, the query [cosco] finds these three events of ships passing with either COSCO containers or ship named COSCO: https://www.camio.com/c/31fcft579lhc.
This has been the goal of many traffic carpool lanes to allow multi person vehicles to pass without payment while single users are charged. It has to work with sports cars and tinted windows along with passenger vans where people count if they are in the back seat.
Good luck!
Perhaps a thermal/visible hybrid camera would be of help here. Sure it might count a dog, miss a baby on a lap (or a propped up Bernie), but what is the required accuracy of the app?
Its true that thermal cameras are particularly poor when it comes to glass, but it might perform well enough to tell a person from a laundry bag by the relative heat and with the visible cameras added info. Watch out for reflections from the camera area itself in the glass as well, as these could overwhelm the true signal.
I know we all want to provide solutions for these types of scenarios, but I feel like sometimes we are offering square wheels for someone with a flat tire (a round one). Just use a guard for three hours. Couldn't be more than $100 / shift.
Just use a guard for three hours. Couldn't be more than $100 / shift.
Perhaps you are referring to M. Silva's post, since I'm unsure where else the three hour spec came from.
If so, you will be happy to know that they never implemented it, and instead ended up using a guard. Also, he is not providing a solution, but rather is politely indicating his skepticism of video analytics for this application.
On the other hand, the OP only said day and night, so...
https://www.technologyreview.com/s/408818/foiling-carpool-lane-cheaters/
That was back in 2007. I wonder if it worked and how much the total cost was.
Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.