Research On Key Technology Of Semi-Active Positioning Based On Image And Wireless Signal

Wireless signal fingerprint positioning such as WiFi has significantly improved indoor positioning coverage,positioning accuracy,and real-time performance,and has been more commonly used in the field of pedestrian navigation.As a result,a large amount of positioning data has been generated.Correlating positioning data and identity can more effectively mine the hidden information in the positioning data(personal preferences,areas of frequent activity,personality characteristics,etc.),which can be widely used in personalized navigation,precise placement of commercial advertisements,accurate market analysis,quick rescue,tracking criminals,etc.According to the source of positioning information,indoor positioning methods can be divided into passive positioning and active positioning.Passive positioning has a wide positioning range and can obtain more positioning data.However,passive positioning can only obtain the Mac address of the mobile device,resulting in a low degree of identity association.In active positioning,the image-based positioning method uses images captured by existing surveillance cameras in public places to obtain rich personal attribute information such as the pedestrian's name,age,gender,and physical condition.However,due to the limited number of surveillance cameras,there are many blind spots where images cannot be collected,which leads to a narrow positioning range and little positioning data.In order to obtain more extensive and accurate identity information,and accurately correlate positioning data and identity information,this paper designs a positioning process based on images and wireless signals for semi-active positioning,which combines active positioning and passive positioning Advantage.The specific process is as follows:First,obtain extensive positioning data through passive positioning.Secondly,obtain rich identity information through active positioning.Finally,through the location trajectory matching,a wide range of location data and rich identity information are associated.This paper mainly completed the following three tasks:1.Aiming at the problem of low accuracy of face recognition,this paper analyzes the advantages and disadvantages of the existing loss function and combines the existing loss function,so that the accuracy of positioning identity is improved.The performance of the combined loss function was tested on the 34-layer and 50-layer neural networks,and the results showed that the average false recognition rate decreased by 10.9%and 7.3%,respectively.When testing on a 100-layer network,except that the recognition rate of the LFW dataset has decreased by 0.07%,the false recognition rates of the other two data sets were reduced by 5.9%and 2.2%,respectively.The experiment proves that the performance of the proposed combined loss function is better than the existing loss function,which significantly improves the accuracy of face recognition.2.Aiming at the problem of low accuracy of face recognition,this paper develops the static loss function of the face recognition neural network into a dynamic loss function,which reduces the overfitting of the training data and further improves the accuracy of positioning identity.The performance of Dyn-arcFace was tested on different layers of neural networks(34,50,100).Compared with the existing loss function,the false recognition rates were reduced by an average of 15.2%,11.0%,and 4.6%,respectively.Compared with the loss function proposed in point 1,the false recognition rates were reduced by an average of 5.0%,10.6%,and 10.6%,respectively.Experiments show that the performance of the proposed dynamic loss function is better than the existing loss function and the loss function proposed in point 1,which significantly improves the accuracy of identity positioning.3.Aiming at the problems of poor wireless positioning accuracy and low matching accuracy of pedestrians and mobile devices caused by unstable wireless signals,this paper builds a neural network model,which uses the spatial distribution correlation of wireless signal strength to modify the wireless signal strength.Simulation experiments show that the average error of the wireless signal strength corrected by the neural network model is reduced by 35.92%.Using the corrected signal to locate in neural networks such as BP,CNN,RBF,and GRNN,the positioning error is reduced by an average of 15.54%.This proves the effectiveness of the correction signal,which is more conducive to obtaining accurate wireless positioning and is more conducive to the matching of pedestrians and mobile devices.