FireBird - the smart camera
The FireBird is a camera combined with an artificial intelligence system, able to prevent fire before it happens.
Description
The innovation we worked out complies under the name “FireBird”, preventing fire before it happens. The FireBird is basically a smart camera, observing the factory (or office) where it is applied to from a bird’s eye view, but with significant hardware features and an artificial intelligence software (AI) operating externally. Camera: - High Definition resolution (fast recognition of possible dangers, but also catching all potentially relevant details for the AI) - Thermal image function (spot unusual heatings) - wide angle objective (broader coverage of the room and less cost as less cameras needed) - SIM card slot/Wi-Fi (Download updates and feeding the AI on Allianz's servers) The gathered data is especially important when it comes to preventing the fire, as technical algorithms analyze and use these to learn what makes FireBird a smart camera. After the installation and first activation, FireBird already knows within the first second where to set a focus. This includes e.g. cables, outlets and the junction box, machines or air conditioners and heaters. Any potential risk area can be added during and after the installation process to provide a 100% secure environment from the first activation. In order to guarantee a frictionless fire-preventing camera, FireBird is able to learn more about its environment with every second it prevents fire and protects the employees. The lens recognizes the movement and learns where passages or tables as well as storages are located. Over the time, it then knows the usual work behavior of the employees as well as the usual heat occurrence e.g. light bulbs, machines or cigarettes, so that the FireBird can recognize patterns which probably lead to fire. This will ensure an even more reliable and safer working day. The technique used for FireBird enables a more precise surveillance of heat arising areas. Furthermore, an accurate localization is enabled in order to alert the fire department and management about the hazard. FireBird is able to differentiate a usual heat (e.g. a light bulb) with an unusual increase of heat (e.g. smouldering). It will consequently alarm the fire department and the management. With the early arriving of the firefighters, they are able to actually prevent a fire or, in worst case, extinguish a fire. Major damages will be then avoided. FireBird will avert worse case scenarios by automatically alarming the fire department. As most of the used hardware is common and already used for surveillance worldwide the product will be affordable also in Turkey. The real innovation hides in the inside: a software that is capable of analyzing videos in HD resolution with an artificial intelligence (AI) for detecting very detailed changes within the surveillance area. The initial set-up of the AI system will not be cheap, but as it can not only be used in small or middle sized factories, the ROI is estimated high - especially when it comes to saving lives. The FireBird learns. It will do so with the new AI that is connected to the surveillance software. Its purpose it to identify all different causes of fire ignition. For this the camera will be trained two times: before delivery to the customer with a stock video database and after mounting with a local video database or through internet connection. At the installation, the Camera is connected to the FireBird software, where the technician will be able to see a live video. Now different zones are drawn simply with a pencil tool in the application. This defines the different zones. For each zone, the algorithm of the surveillance software can be changed. Samples for zones from which the technician can choose can be already implemented. For example, the desks of the workers in a factory will be the first zone and a place where cotton is stored will be the second one. Whereas the work zone has heat sources like lamps and machines the storage area might have none. Therefore, the algorithm will react with the thermal camera highly sensitive to heat development in the storage. In the work zone the algorithm will learn the amount of heat for every source and put focus on differentiating the sources. This makes it possible to recognize smoking workers for example. Software: The stock version contains a big number of videos from different perspectives, angles and causes of fire ignition. The software will then remember situations and how they looked like before a fire occurred. Every test video’s alerts are tagged with a time marker. If the algorithm fails to detect the source of fire it will automatically repeat the detection and adapt until the result is correct. This training ensures that no false alerts occur once the camera was mounted. It also learns how objects look from different perspectives such as boilers or cigarettes. The second training time will be initiated by the technician but completed by the AI itself. The software will detect the surveillance zone for a few days to learn regular movements of objects and workers but also measure all heat sources like lamps and machines to identify the normal heat levels. After this training, the software will be able to reveal critical heat levels. To minimize false alerts the smart algorithm will compare the detection results of both cameras and then decide whether there is a real danger or not. This solution was created and developed in a team of 5 students and was part of an academic lecture.
References
Surveillance cameras are omnipresent and the technology regarding smart data analytics become more and more important.