A large proportion of Video Analytics Projects fail usually resulting in finger pointing by everyone involved.
In our experience a majority of project failures result from two main issues.
1. The Wrong Software
2. Incorrect Cameras and Camera Placement.
As the terms Artificial Intelligence and Video Analytics become more popularly recognized large numbers of suppliers have entered the market offering everything from Face Recognition to Behaviour Analysis. The technology is taught at many Universities and many individuals have felt that they could use this information to build and market a system.
Unfortunately, most of them have not realized that for systems to be useful they have to be robust. They have to be able to operate in real world environments which are often crowded and complex. They have to cope with varying environmental conditions.
There are, however, no universal tests that can help to differentiate simple and complex situations. The available tests such as iLIDS and NIST are designed for relatively simple situations.
An airport in the Middle East released a tender for detecting abandoned objects. Many people bid and there was a significant price difference between the suppliers. They purchased the least expensive system and the system just could not detect abandoned objects. On complaining to the supplier they were asked to do a test – empty the area, leave a bag and the system would detect it.
The airport complained that this was not a realistic scenario as airports are invariably crowded and one needs to detect the abandoned bag in a crowd. However, they found that the supplier had legally met their requirements because the airport had not specifically asked for the detection to be performed in a crowd.
If in its requirements the airport had said that the system needed to “detect bags in a crowd even when the bag is obscured for significant periods of time” then they would have eliminated all the shoddy suppliers and got a system that would have actually addressed their problem.
• Ensure that the system is explicitly armed with an artificial intelligence based NAMS system (Nuisance Alarm Minimization System) for minimizing false alarms.
• Check and get evidence that the system can work in real life crowded environments.