Whilst many traffic surveys can be carried out using either sensors such as ATCs or infrared pedestrian counters or computer vision technologies, there is still a very large subset of tasks which all but require human intelligence.
Object tracking and classification are difficult machine learning problems at the best of times, but when dealing with temporary camera installations, these problems are magnified.
Depending on the competence of site technicians, temporary installations can encounter any of the following problems, and these can have a dramatic effect on the performance of computer vision solutions.
- Low Height Mounting - This increases the occlusion of vehicles/people by each other and other objects
- Insecure Mounting - May introduce unwanted movements and shaking
- Poor Positioning - May introduce glare, overexposure, poor contrast, or in extreme cases important parts of the survey scene are missed or unclear
- Poor Zoom/Focus - Reduces clarity of the video
- Poor Lighting - Reduces clarity of the video
In permanent camera installations making use of computer vision technology, the 3D environment may be modelled in advance, to improve the accuracy of data collection, but temporary installations can very rarely afford that luxury.
Site technicians are often under pressure to complete the set-up of several cameras according to strict time schedules, and quality frequently suffers.
Fortunately, unlike software algorithms, humans are very forgiving of the impact these problems have on the resulting video footage, and in all but the worst cases can still extract accurate data.
Video Datapad, is carefully designed to serve a supporting role, and allow video enumeration work to be outsourced efficiently, and for many types of survey to be easily supported.
The service is currently in beta-testing, but please let us know if you would like to try it, or discuss it's potential application in your organisation.
