Research On Optimization Algorithm For Inertial Navigation Positioning Of Firefighters Based On Random Forest Combined With Map Aided

Building fire is one of the main disasters that occur frequently every year in China,which seriously threatens the safety of life and property of the public.When a fire occurs inside the building,firefighters usually need to enter the fire scene for fire fighting and rescue.However,the complexity of the building structure and the harshness of the fire environment often cause firefighters to lost in the fire and can not determine their location,and the outdoor rescue command center can not also determine the true location of the firefighters.The importance of indoor positioning is increasingly prominent in the field of intelligent fire protection,which is of great significance in ensuring the life safety of firefighters and improving the efficiency of rescue.In the face of indoor fire environment,conventional positioning technology such as GPS,BDS and WIFI are no longer applicable.With the rapid development of MEMS inertial measurement unit(IMU),MEMS inertial navigation has become the most suitable positioning technology for emergency in fire environment.Due to the low precision of the MEMS-IMU,the positioning error diverges rapidly with time and is larger under the variable gait of pedestrians.Based on the MEMS-IMU BWT901 BCL,this thesis optimizes distance error and heading error of the pedestrian inertial positioning respectively,and studies the inertial positioning optimization algorithm suitable for firefighters to provide technical support for future firefighter positioning and the realization of intelligent firefighting.For the correction of distance error,considering the variable working conditions of the firefighter’s gait,the six gait patterns which are standing,walking,jogging,running,going upstairs and going downstairs are established based on zero velocity detection of generalized likelihood rate test(GLRT)with optimal robustness,and the values of the GLRT parameters are determined.Using machine learning for reference,this thsis selects a random forest to built a random forest gait classification model,and combines with GLRT zero velocity detection to form a random forest zero velocity detection model(RFZVD).Based on the traditional SINS positioning distance optimization algorithm(SEZ),this thesis proposes an inertial navigation positioning distance optimization algorithm based on random forest zero velocity detection(SREZ).Through the inertial data acquisition experiment in rectangular route,the algorithm is verified by MATLAB.The conclusions obtained are as follows: the random forest gait classification model established in this thesis has good ability of gait classification,and the overall accuracy can reach 97%.The RFZVD proposed on this basis has better zero velocity detection ability than the traditional zero velocity detection model.The SREZ algorithm proposed in the scenario of firefighter’s changeable gait has better positioning accuracy than the SEZ algorithm,and distance error of the overall positioning is smaller,but there is still heading error.For the correction of heading error,this thesis introduces particle filter combined with map aided model.However,this model needs to consider the influence of systematic observation in the particle weight update stage,which make it no longer be applicable when heading error of the original trajectory is large.Therefore,this thesis improves the particle weight update and resampling rules,and establishes an improvement of particle filter combined with map aided model suitable for this thesis,so that the pedestrian positioning trajectory is more reasonable and accurate,and finally verifies it by calculation examples based on MATLAB.The conclusions obtained are as follows: the example results show that,compared with traditional particle filter combined with map aided model(TPFMA),the improvement of particle filter combined with map aided model(IPFMA)established in this thesis has a better heading optimization effect,and can effectively optimize the original trajectory with different heading error sizes to the actual position,it has better adaptability and robustness to different working conditions of heading error.Combining the above two algorithm models through a cascade structure,this thesis proposes an optimization algorithm of inertial positioning based on random forest combined with map aided(SREZPPM).The combined optimization algorithm consists two level models,the bottom model is a distance optimization algorithm of inertial navigation positioning based on random forest zero velocity detection(SREZ),the upper model is a heading optimization algorithm of inertial positioning based on particle filter combined with map aided(PPM).A dormitory building is selected as the actual building model,and inertial data acquisition is carried out by using BWT901 BCL according to the pre-defined route,and finally the algorithm is verified based on MATLAB.The conclusions obtained are as follows: under the pedestrian walking gait,the positioning accuracy of the SEZ algorithm and the SREZ algorithm are comparable,and the distance error is relatively small,but heading error is still exist,while the heading error of the SREZPPM algorithm is the smallest,and the final overall positioning error is the smallest;under the pedestrian random gait,the positioning accuracy of the SEZ algorithm is the worst,and the SREZ algorithm is the second,while the distance optimization and heading optimization of the SREZPPM algorithm are the best,and the final overall positioning error is the smallest,basically within 0.6m.Regardless of the gait state of the positioning personnel,the dormitory building positioning experiments show that the overall positioning accuracy of the SREZ algorithm and SREZPPM algorithm in this thesis are significantly better than that of the traditional SEZ algorithm,and they are robust.