Published on

Deep Neural networks based Fire Preventing System

Introducing a DNN powered fire warning system that "sees", analyzes and detect fire hazard, to prevent fire incidents to happen

Description

Neural networks are a set of algorithms, modeled loosely after the human brain, that are designed to recognize patterns. They interpret sensory data through a kind of machine perception, labeling or clustering raw input. The patterns they recognize are numerical, contained in vectors, into which all real-world data, be it images, sound, text or time series, must be translated. Such systems are usually based on strong servers, usually reside on cloud mega networks, and require large CPU power to analyze. However, recently there are several development in the embedded world which allows integrated DNN based mechanisms inside small silicon components. The result is the ability to create low price DNN nodes with Computer Vision ability along with specific training to detect specific images, patterns, and connections. Based upon these new technologies, the idea is to create a DNN based sensor connected to a standard mini-wide-angle camera (or to a larger spectrum camera with infra-red vision ability) and a communication media (i.e. WiFi, LTE) to build independent watchers. These watchers shall be able to "see" their surroundings, to identify objects and to build pattern of these objects to understand what is happening around them. Based upon this analysis, the watcher shall be able to warn on dangerous behavior that may lead to fire related incidents (or to other hazards, depend on the DNN training process). For example, imagine Joe, a worker who performs ironing process to clothes. During his work, he received an important call from his wife, which made him forget the iron machine on the cloth. Such behavior may lead to fire in a few minutes. The DNN watcher shall detect the iron machine, detect that it is placed on a cloth and not in a secure location, and after several seconds with no move - shall detect the fire hazard and start warning about fire danger. Other improvement to this system may be smoke detection sensor, which expand the ability of the DNN watcher to better understand its surrounding environment conditions.

References

Introduction to Computer vision and DNN: https://venturebeat.com/2017/01/29/8-cool-new-ways-computer-vision-is-changing-everything/ Existing Embedded systems sensors and AI components: https://www.cmu.edu/nanotechnology-forum/Forum_14/Korea_Presentation/9_Yoo%20Slide.pdf FPGA solutions related to DNN: https://www.hpcwire.com/2016/03/17/fpga-opencl-solution-released-deep-learning/

Evaluations