Research On Mine Personnel Target Positioning Technology Based On Compressed Sensing And 3D Image

With the high-efficient mining of coal resources nowadays,the mine safety of coal is always the focus concerned by people all over the country.In which the personal safety of mine working personnel is the key issue concerned by the whole coal industry.With the continually developing of technology in this period,the coal mine safety system is more and more completed and more and more researching works are focusing on mine personnel positioning and identity detection to optimize coal mine safety system.Based on this topic and relying on the National Key Research and Development Program of China(NO.2016YFC0801800)and the National Natural Science Foundation of China(NO.51674269),as well as combined existing technologies such as compressed sensing,fingerprint database positioning,wavelet transform,3D face image recognition,etc.,mine personnel positioning systems and methods are operated deep research and four mine personnel positioning methods and a 3D miner image human-face nose tip detection method are proposed,the main research and innovations are as below:(1)Facing to the situations of low target positioning accuracy,poor positioning real-time performance and heavy data collection workload in current domestic mines,this paper brings up a method to establish fingerprint database based on the distributed compressed sensing and realizes mine personnel positioning.The method collects a small amount of fingerprint information along the tunnel in the offline phase to reconstruct the fingerprint information of mine target fingerprint database by high probability.So as to achieve the purpose of reducing data collection workload and improving work efficiency.In the online phase,the method only needs to obtain ID information of reference nodes and the real-time distance measurement values of target node from reference nodes in some time.According to model matching method,the estimated distance values from the target node to reference nodes in the time can be obtained.The method guarantees the positioning accuracy and timeliness.Based on this,an Improved Compressed Sampling Modifying Matching Pursuit algorithm is put forward to reconstruct fingerprint information,which can effectively shorten the reconstruction time.(2)This paper discusses the sparse characteristic of target location in the space domain and certain time,which could operate mine target wireless positioning in use of sparse sampling compression characteristics of compressed sensing theory.Using compressed sensing positioning model could efficiently reduce the amount of online measurements,meanwhile it can realize high positioning accuracy.The traditional positioning methods based on compressed sensing are mostly ranging,which is not suitable for low-loss wireless sensor networks that are limited in energy,so this paper proposes a mine target positioning method based on non-ranging compressed sensing.In this method a non-ranging compressed sensing positioning model is designed according to connecting information between sensor nodes and target nodes,and a fingerprint database is built for positioning area,which solves the problems of large node layout,low accuracy and delay of non-ranging positioning methods.(3)The signal energy measurement method of traditional fingerprint database based on energy attenuation model on every positioning grid is tedious and complex,even accomplished it divides the positioning grids into large parts and the amount of grids that need to be measured is small,which severely affects the positioning accuracy of mine target.Based on energy attenuation characteristics of electromagnetic signal propagation,combined with compressed sensing model,a mine target positioning method based on energy attenuation matrix is proposed.By constructing the fingerprint database corresponding to the grid position coordinates and the grid numbers,the measuring matrix and the sparse matrix based on the energy attenuation model,taking use of improving Greedy Matching Pursuit algorithm to reconstruct the target position signal.The method simplifies the positioning model,and not only significantly reduces the task load of constructing database,but also lightens the power consumption of the positioning system and guarantees the accuracy of positioning results.(4)In the environment of coal mine underground tunnel,it is difficult to establish a unified electromagnetic propagation model because of the factors that multi-path effect and electromagnetic wave propagation loss are numerous and complicated,in addition,the trilateral positioning of electromagnetic wave ranging is very demanding on the hardware system.In order to solve the problems,a distributed compressed sensing multi-target positioning method based on wavelet transform that satisfies the underground tunnel is proposed.The method takes fully use of spatial correlation information of mine tunnel nodes and transforms the target positioning problem into joint recovery problem of signal group data.Simultaneously,through non-cooperating way and a few number of sample values it could accomplish target positioning,which efficiently lowers the extra power consumption caused by exchanging data between nodes and satisfies the design concept of low power consumption and lightweight of underground tunnel hardware.(5)This paper analyzes the shortcomings of 2D image monitoring technology of mine and discusses the advantages of 3D face recognition technology applied in coal mine for face recognition and personnel position monitoring.For the importance of 3D face feature extraction,according to the characteristics that miner usually wear safety helmet with miner's lamp and glasses,the paper proposes a 3D miner image human-face nose tip testing method based on threshold segmentation.The method would find out the locations of miner's lamp and glasses through 3D Hough transform.Then using region segmentation method removes the occlusions of helmet,miner's lamp and glasses,meanwhile eliminates the impact of occlusions on HK classifier and enhances the success rate of nose tip detection.The algorithm is simple because it does not need model and training.