Research On Unsafe Behavior Recognition Method Of Miners In Cage Based On Kinect

The characteristics of China’s energy resources determine that coal is still the largest energy source,safeguarding coal mine safety production is the top priority of industrial safety.Human unsafe behavior is the most important cause of safety accidents.The traditional prevention of miners’ unsafe behaviors is mainly through safety training and field supervision,with low efficiency and reliability.How to effectively and reliably screen miners’ unsafe behavior with the help of information technology to strengthen accident prevention and management is very important,which is of great significance to reduce the incidence of coal mine accidents.The auxiliary shaft is the "throat" position in the coal mine,and the deep underground mining has become a trend,which puts forward higher requirements for the protection of the safety problems of the auxiliary shaft lifting system.This thesis selects the research scene in cage of auxiliary shaft and uses Microsoft Kinect Body Sensing sensor to study the behaviors of miners in the cage of the auxiliary shaft and to judge the behavior nature of the miners by combining the image information of the auxiliary shaft safety door,which is of great significance to reduce the occurrence.This thesis studies the dynamic behavior and interaction behavior of miners in cage.In the aspect of the dynamic behavior representation of miners,this thesis studies it on the basis of the representation of the static posture of the human body,and puts forward a method of dynamic behavior representation of miners based on the characteristics of time pyramid of the maximum information joint angle sequence,which overcomes the limitation of traditional behavior characteristics and has stronger adaptability to represent the action sequences with large difference in length.In the aspect of the expression of the interaction behavior of miners,In order to improve the accuracy and robustness of miners’ interactive behavior representation method,this thesis extracts the feature of joint distance based on keyframes,and on this basis,the features of plane distance and vertical plane distance are fused.In order to prove the effectiveness of the method of miner behavior representation in this thesis,SVM modeling is carried out by self-built cage miner behavior dataset,the experiments show that the accuracy of dynamic behavior recognition of miners is 94.8%,the accuracy of the identification of miners’ interaction behavior is 84.6%.This thesis also proves that the proposed method of dynamic behavior representation and recognition of miners is applicability on MSR Daily Activity 3D dataset.In order to further improve the accuracy of the identification of miners’ interaction behavior and to improve the robustness of the interactive behavior Classification model,this thesis proposes a method of miners’ interactive behavior recognition based on decision-level fusion,and the accuracy of the recognition of miners’ interaction behavior reaches 88.3%.Considering that the unsafe behavior of the miners in cage is related to the state of the auxiliary shaft safety door,this thesis study the identification of the auxiliary shaft safety door state based on the image expansion.Through the field investigation of several coal mines and access to relevant regulations,this thesis constructs the classification template of miners’ behavior nature in cage,and with the help of test samples,this model is used to identify the behavior of miners and the state of auxiliary shaft safety door,then combine the two codes,and finally match the classification template to identify the miners’ behavior nature.The final accuracy of the identification is 94.2%,which shows that this method has a good identification effect for miners’ unsafe behaviors in cage of auxiliary shaft.