Deep Learning applied in Incident Management
The first and most simple subset of Artificial Intelligence, Machine Learning, has been integrated into Citilog’s products for a long time. The more complex and resource demanding Deep Learning is now a reality with CT-ADL (Citilog Applied Deep Learning). The uniqueness of CT-ADL is the ability to run a specifically developed neural network targeting at eliminating false positive detections while maintaining the increase of hardware requirement to a very limited level: 1 standard additional GPU card can cope with 100 cameras.
CT-ADL: Efficiently eliminating False positive for open roads Incident Management solutions.
Apply Deep Learning to solve a real business case
Theoretically Deep Learning technology can go a long way in improving video analytics performances. However, applying deep Learning to solve an actual business case is more difficult than it sounds. It requires to clearly identify the issue and to build the required dataset to train a neural network towards solving this issue. CT-ADL is the result of this process, targeting to eliminate the false positive detections of the Incident Management solutions using a specifically designed network trained with the vast dataset available to Citilog through 20 years of experience in this specific field.
10 times less false alarms
The results, gathered from multiple sites already benefiting of CT-ADL, show that the amount of false positive detection historically due to shadows, rain patches, bad weather conditions, is divided by a factor 10 compared to traditional analytics. This improvement, makes CT-ADL an operational Incident Management solution for bridges, ring roads and highways even for large number of cameras.
‘The future of Applied Deep Learning in mobility: detection of vulnerable users’
The future of Applied Deep Learning in mobility: detection of vulnerable users
This technology will of course contribute even more to the next generations of analytics for improving mobility in Smart Cities. The abilities and performances of Deep Learning in object recognition and classification, combined with a new generation of cameras featuring DL capabilities on the edge, opens new opportunities for converting video cameras into smart sensors, source of both high-quality videos and valuable and actionable data.