Research On Key Technology Of Crowdsensing Based Mobile Location Recognition

With the development of mobile intelligent terminal,it is more and more convenient and faster for people to obtain and share location information,the contents of which transform from text to image,text and sensory data of in?telligent ends,including various kinds of location-related data.It is not only capable of obtaining the physical coordinates of mobile users,but also recog-nizing the logical location semantic information of places that users focus on,to make full use of the rich data sufficiently,which is significantly meaningful for implementing various location based services.Meanwhile,as a new type of sensing paradigm,crowdsensing can enable the users to collect large scale and complex data consciously or unconsciously,enjoying the benefits of low-cost,high flexibility,and widly-moving scopes etc.Therefore,the thesis focuses on utilizing crowdsensing paradigm to collect location fingerprints,and researches the key technologies of mobile location recognition.From three perspectives(i.e.,location fingerprint database construction,mobile location recognition,fingerprint database maintenance),we propose corresponding methods and key technologies and construct a crowdsensing based logical location recognition system testifying platform,with the research goal of data collection efficiency,location recognition accuracy and low-cost of system deployment.Specifically,the main contributions of the thesis are as follows:(1)High-efficiency data collection oriented crowdsensing task allocation method.Constructing location fingerprint database needs to allocate tasks and plan routes for the participants who take part in data collection,and the tasks have time-sensitivity and heterogeneity.According to Cannikin Law,the time of performing all tasks is determined by the participant who completes the tasks at last.We set the optimization goal of minimizing the longest time consump-tion in task allocation.For this goal,we propose memetic based bidirectional variant neighborhood search algorithm,in which heuristic strategy is designed to initiate the allocated paths,and design the bidirectional variant neighborhood search structure to adjust tasks on the border.Meanwhile,for the variability of the participants,we design the task reallocation mode and a light-weight alter-native algorithm.Through experimental verification,this method improves the efficiency of performing the task within scene of heterogeneous task distribu-tion.(2)Rich fingerprint based mobile location recognition method.In addi-tion to images and location labels,the rich fingerprint data includes other rich sensory data.In order to utilize the rich fingerprint to implement more accurate mobile location recognition.We first propose the object's coordinate centric fingerprint searching mechanism,which utilize coordinate matrix transfer of ac-celerometer to obtain tilt angle to calculate the object's coordinate,then ensure the accurate fingerprint search space and build the sub-area overlapped non-visual index structure;After that,we extract all the ORB(Oriented FAST and Rotated BRIEF)features in images of fingerprint database,use hierarchic clus-ter building vocabulary tree method to train codebook,and build visual word based inverse-order index structure to reduce unnecessary fingerprint match.Through experimental verification,this method is capable of improving the ac-curacy of location recognition.(3)Simplified fingerprint database oriented high-quality fingerprint selec-tion method.Because the crowdsensed fingerprint database has too many re-dundancy data,and the quality of the sensed fingerprints may be also not good,it may waste valuable hardware resource consumption at the mobile terminal,which needs us select high-quality fingerprints to build simplified fingerprint database to implement low-cost system deployment.For one thing,accord-ing to the principle of fingerprint database should have special diversity,we propose self adaptive special clustering algorithm to hierarchically cluster the fingerprints of same object by K-means,to ensure the fingerprints have multi-ple diversity such as directions,angles and distances.For another,according to the principle of selected high-quality fingerprints should have special saliency,we propose the concept of Common Salient Feature(CSF)and design Self-Adaptive Clustering based Common Salient Feature Detection(SAC-CSFD)al-gorithm to select high-quality fingerprint,which utilize Locality Sensitive Hash(LSH)technology to qualify the SURF features in the image database into hash codes,and select high-quality fingerprint through evaluating the number of CSF in the images.Through the experimental verification,this method implements high-quality fingerprints under the premise of accuracy.(4)The design and implement of mobile location recognition system based on crowdsensing.To testify the feasibility and usefulness of the system,we design the crowdsensing based mobile location recognition system and imple-ment several functional modules,including fingerprint collection,task alloca-tion,fingerprint selection and location recognition.In summary,we propose a series of solutions and methods from three per-spectives(fingerprint database construction,mobile location recognition and fingerprint database maintenance),implement a mobile location recognition system based on crowdsensing,and testify their effectiveness,which can pro-vide important theoretical and technical support for crowdsensing based mobile location recognition.