Intelligent Positioning Method Of Indoor Flame In Video Image Based On Improved Yolo-v3

With the rapid development of society and economy,the urban population has gathered regionally,and the indoor population density has increased accordingly,which will increase the hidden danger of indoor fire.So,it is necessary for us to find the indoor flame as soon as possible,determine the flame position,and then extinguish it to reduce the loss and reduce the casualties.Therefore,the rapid identification and precise positioning of the indoor flame is of great significance.Among all the indoor flame recognition and positioning methods,the video-based monitoring is the most common method,but the existing researches on fire recognition and positioning based on video images have problems such as slow recognition rate,low accuracy.In addition,they are impossible to perform accurate positioning based on a single fixed camera indoors.In response to the above problems,we conduct research on the real-time and efficient flame detection method based on video images and the intelligent positioning method of indoor flame based on the spatial semantic constraint rules.We select case scenarios and conduct corresponding experimental analysis.The main work and results are summarized as follows:(1)We proposed a real-time and efficient flame detection method based on video images using the improved Yolo-v3 algorithm.Firstly,we give out the general idea of flame detection and analyze the characteristics of the Yolo-v3 algorithm.Then we improved the Yolo-v3 algorithm from two aspects: improving the multi-scale detection network and optimizing the prior frame setting.Combining the brightness difference of the target area pixels between adjacent frames of the video,the flame flicker feature was extracted.Finally,we designed a process for real-time and efficient flame detection based on video images using the improved Yolo-v3 algorithm,which improved the accuracy and speed of flame detection;(2)We proposed an intelligent indoor flame positioning method based on spatial semantic constraint rules.Firstly,we designed the general idea of intelligent positioning of indoor flames based on spatial semantic constraint rules,and discussed the method of building indoor scene models under spatial semantic constraints.Then we analyzed the indoor target positioning method based on video images,and studied the two-dimensional and three-dimensional space position conversion method based on monocular vision image and the flame space position determination method.Finally,we designed an intelligent indoor flame positioning method based on spatial semantic constraints and video images;(3)Scene construction and experimental case analysis.Based on the above research results,we selected the case scene,constructed the case scene flame video data set,established the indoor three-dimensional basic scene,and carried out the flame recognition and spatial positioning experimental analysis.The experimental results show that the average accuracy of our flame detection method can reach 98.5%,the false detection rate is as low as 2.3%,and the average detection rate is 52 frames per second,which has better performance in terms of accuracy and speed.The average positioning error of our positioning method is 2.66 cm,which is much smaller than the flame radius.This result can fully meet the flame positioning accuracy requirements.