| With an increasing need for location services,wireless positioning technology has advanced rapidly.At the moment,wireless positioning technology is mostly used as locate target carrying equipment,such as satellite positioning,base station placement,and so on.However,in some applications,such as fire rescue,intrusion detection,and others,the target may not carry the device,necessitating the use of target device-free localization(DFL)technology.Because Wi-Fi signals are ubiquitous,no specific equipment is required for localization,making DFL based on Wi-Fi signals an important research direction.Although the indoor location technology based on Wi-Fi signals is low-cost and simple to deploy,the Wi-Fi signal is unreliable and easily impacted by the surroundings.As a result,achieving high-precision DFL based on WiFi signals remains a significant challenge.In this study,the DFL technique based on widely available and simple-toimplement received signal strength(RSS)is examined in order to obtain DFL with high accuracy and low error,and the related work is as follows:Firstly,after evaluating the benefits and drawbacks of several sensor network configurations,a mobile transmitting sensor arrangement is chosen in this paper to maintain the same number of wireless links while lowering sensor costs.The transmitting sensor node in this sensor model is a moveable sensor,which moves continually at different locations to send Wi-Fi signals,while the rest of the receiving sensors receive RSS cyclically as the data set for localization to achieve DFL.Then,to solve the problem of localization data potentially overlapping in lowdimensional space during processing,this paper incorporates kernel functions into sparse coding based on DFL and combines the two to propose a new algorithm.To begin,the kernel transformation is applied to the original data to increase the dimensionality of the data,resulting in greater separability.Taking into account the computational complexity,the high-dimensional kernel space features are then extracted using the feature extraction method,resulting in dimensionality reduction.Finally,the optimization problem is solved by sparse representation.This algorithm compares different algorithms when different noises are added,and the experimental results show that this algorithm has better localization performance than other algorithms.Finally,to solve the issue that RSS is more subject to environmental changes and that the sparse coding of fixed dictionaries may influence DFL performance.A localization approach based on kernel function and dictionary learning is proposed in this study.While enhancing RSS data separability,it can also train dictionaries that are more flexible and suitable to volatile RSS.The kernel transformation is applied to the RSS data first,followed by the kernel dictionary learning on the kernel converted data.The kernel dictionary learning process is separated into two stages: kernel sparse coding and kernel dictionary updating;which are alternated to produce a pseudo-dictionary,after which the target position is predicted using the generated pseudo-dictionary.The experimental results show that the algorithm outperforms the general dictionary learning algorithm and the sparse representation algorithm in terms of localization. |
