Research Of Indoor Location Algorithm Based On CNN Activity Recognition Aided PDR

In recent years,Location Based Services(LBS)has been a hot topic of research.In outdoor positioning,under the background of GNSS,various location-based services have emerged,which has brought convenience to people's daily routine.Although GNSS has a unique advantage in the outdoor,its signal is highly susceptible to edifices,dense foliage and etc,and thus lead to multipath effects,resulting in a sharp drop in positioning accuracy.The positioning errors generated in these places and indoors are extremely large and the expected positioning effects are not available.While people's indoor work and live have always been the major part of their daily routine,for which to seek an indoor positioning technology with the same high precision is especially important.This paper analyzes and studies the existing indoor positioning technology,combines intelligent mobile devices and related machine learning algorithms as the research direction,and improves the defects inherent in the existing PDR technology,that is,the problem of the accuracy caused by the accumulated error,and conduct data analysis on the daily behavior of pedestrians' indoor activities,proposing a method based on CNN activity recognition to assist PDR indoor positioning.The specific content includes:(1)The researches of recognition algorithm based on CNN activity were accomplished.Data analysis was carried out on daily activities such as pedestrians' walking,left-turn,and right-turn,and a CNN activity recognition model was established to realize the recognition of three kinds of daily activities of pedestrians.(2)The startup algorithm of the CNN activity recognition model was studied.By analyzing the variation of signal strength error,the signal strength change threshold is obtained,and the AP signal strength at each feature point is detected,and the feature point fingerprint database is established.Based on that,a feature point fingerprint algorithm is proposed to realize the switch between PDR location and CNN activity recognition.(3)Systematically design the positioning algorithm based on CNN activity recognition aided PDR.The system is divided into mobile client and server.The users' interaction and output of positioning result have been accessed in the mobileclient.In the meanwhile,data analysis,data preprocessing,the construction of CNN activity recognition model and positioning operations are major tasks of the server.(4)A comparative study on the sensor acquisition of data denoising algorithm was carried out.During pedestrians' movements,the data collected by the intelligent mobile device sensor is denoised,and the methods of denoising are diverse.Thus the best denoising method is obtained by comparison of the effects of these diverse methods and then used for data denoising.In conclusion,experiments show that the CNN activity recognition aided PDR positioning algorithm proposed in this paper has better error correction effect at the feature points,and the higher indoor positioning accuracy has been obtained from this algorithm,meanwhile,the reduction of the accumulative error of PDR positioning has been achieved.The high availability and high precision requirements of indoor positioning in terms of practicality and positioning performance have also been satisfied.