A sense of security is extremely important for passengers, especially during night time. Providing them a secure travel experience is the goal of public transport providers. Therefore it is essential to recognise security and safety hazards as early as possible and to handle them in the most effective way.The research project InREAKT deals with the development of a system that will detect critical incidents automatically, and initiate a quick and assisted response.
Incident Management
What is being developed in the research project InREAKT
Early detection
Live video streams are processed by neural networks specially trained to recognise these situations. Of course, not all sitting objects have actually been forgotten. When properly trained, however, this technology will know the difference with more certainty. Beyond a given level of probability, a threshold, the algorithms will trigger a warning report. Further individual intelligent cameras on the bus may confirm this warning, increasing the reliability of the observation. In the same way, later reports may strengthen or alter the severity and even the type of the initial report.
Eyes and ears
Human input
When time is of the essence, it is vital to quickly share precise information and to receive valuable feedback. This is why the exchange of information, to and from local staff, auxiliary services, passengers and bystanders is also welcome. Although professional and civilian sources should be evaluated differently, almost all human input should receive a high level of credibility, such as incoming emergency calls. Unfortunately, human input is often unstructured and therefore difficult to process automatically. A dedicated app largely helps to overcome this: one for professionals and one for civilians. The difference lies in the accepted levels of access, trust, data privacy and protection.
Assistive dispatching
Self learning
This is where the log file comes in. You must learn from your mistakes, so you can continually improve your team! The system’s neural networks are trained to match initial incident reports with incidents that have been handled effectively. Feedback to adjust the parameters of the sensors is one area of self-learning. The modifications to the suggested incident handling procedures is a second.
Outlook
Contact
Subscribe now
This article was published in our customer magazine INITiative, issue 2/2015. If you want to receive the magazine regularly per email, we kindly ask you to subscribe.