We started this project as we couldn’t locate any cameras system around the world that was rugged enough to handle vehicle and on site conditions, be able to recognize objects and interact in real time with a REST server and be at a reasonable price
There were four main purposes behind this product requirements:
1...Vehicle safety - we need to be able to record 360 degrees around the vehicle - every day.
2. Object Recognition - if a customer puts the wrong item into their compost bin - we should be able to recognize it and report it in real time..
3. What IF Actions - these relate to preprogrammed requests - like we expect a bin to be at this geofence location (if no bin lift) then send photos to server, as bin not presented. Another would be to every hour check the contents or a large container and send fill level and contaminated contents to the server.
4. Site Monitoring - we get a large request for this as we have many recycling companies who have collection sites which need monitoring with automated actions.
To deal with these 4 main business requirements we have had design the camera system from the bottom up. Unlike conventional camera systems that operate through a DVR we decide to design every camera with its own micro PC meaning they could think and interact with onlines services independently.
The main hurdles we faced were environmental conditions as we knew the PCBs and the technology were scalable and adaptable with software but trees hitting, heavy rain and extreme sun were a serious concern. To deal with these we purchased our own refuse vehicle and setup testing rigs around our yard. Firstly we spent many months designing moulds (casing) that could withstand a direct tree hit and slide away from the vehicle but at the same operate correctly and look aesthetically pleasing. To achieve this we had to design a back plate radiator, unbreakable glass dome and flexible wiring lume. Eventually we go our device to withstand 72 hours in 1foot of water and still operate correct (as close to IP69X as you can get)
Next step was to get the 220 degree fisheye camera to recognize objects in details from a distance of 4 - 5 meters away. To achieve this we had to dig deep into the Python and Java code to push the HD camera to its max and then to work backward to a normal working environment in order to read vehicle number plates for 3.4 meters away
Currently we have the camera system installed on our refuse test truck and will be doing early customer trials in September 2019.