50% is the reduction of traffic violations in the city of Zagreb after the municipality installed video analytics (VA) solution. The local authorities have tripled their revenue from traffic violations utilizing such tool.
Deeper into the city of Zagreb video analytics story
The capital of Croatia has 800,000 residents and a bustling city center. With heavy traffic of vehicles and pedestrians combined with pretty specific local culture, a lot of drivers park their cars right on the footpaths.
That’s why the local authorities decided to bet on VA. Of the 225 cameras deployed at critical intersections and traffic points around Zagreb, 86 of them were enabled with real-time event detection capabilities and configured with “stopped vehicle” rule. Crucial for the final result was the integration with the municipality Security Center, the ERP system and with the national car registry database. This way, Zagreb capitalized the surveillance system investment while increasing road safety.
Machine learning and video analytics
You`ve seen this on television – crime has been committed and officers are staring at security footage to see if any of it was caught on camera. In real life, however, analyzing huge quantities of video data is a task that’s rarely accomplished effectively by human operators. There is too much data, hence video, to be observed by any staff and the cost of the labor would be really high. This is the spot where machine learning and video analytics are really effective, says Bernard Marr in a recent publication.
He adds that VA can be used for face recognition, behavior analysis, and situational analysis. For instance, you might count road traffic, to detect cars in need of assistance, illegal U-turns, or to track people moving in your store, to identify possible theft or entrance in restriction areas.
There is a lot more – VA is capable of detecting suspicious objects left behind at airports. It might be used as a counter for parking lots occupancy or to track suspicious people at the parking lot, unauthorized exits, trucks moving in the wrong direction or to track people in heavy weather as heavy snowfall, storm or rain.
Is there a match between predictive analytics and video?
MIT team created a new algorithm in which computer can predict human actions from what the people are doing in the seconds before. It was created at the Computer Science and Artificial Intelligence Laboratory where researchers fed the program with 600 hours of YouTube videos and shows to see if it could learn about and predict certain human interactions like hugs, kisses, high-fives, and handshakes.
To test the algorithm, they showed it a video of people who are seconds away from doing one of the aforementioned interactions. The computer showed 43% success rate, compared to 71 reached by actual humans. The MIT team says that the algorithm will be much more successful if it consumes more video data than the 600 hours used for the experiment.