Research on real-time fire detection and locating for automotive firefighting robot in factories based on Convolutional Neural Network

Abstract

Automotive fire robots that are used in factories can carry out diverse operations in regard to patrolling, fire detection, and programmed fire rescue. An accurate detection of fire sources in factories is significantly crucial for unmanned firefighting robot in terms of building a reliable sensor system. An approach proposed by this paper to recognize the color and dynamic shape of varying flames based on HSV color algorithms and Convolutional Neural Network. As a comparison to traditional RGB image processing, this approach is more efficient in isolating colors in environment and more adaptive to a fire site that includes multiple noise factors. The research in this paper uses image processing algorithms that is trained by CNN to detect flames in simulated factory environments, followed by a HSV color locating algorithm to compute the coordinates of target fire to perform inverse kinematic analysis on an unmanned firefighting robot.