Individual target tracking method based on user behaviors

Abstract

The invention provides a user behavior-based individual target tracking method, which comprises the following steps of: acquiring signaling data, and constructing a trip link database according to the signaling data; key clues of individual targets are obtained, the key clues are matched according to the trip link database, and candidate target digital identity information is obtained; acquiring the digital identity information of the candidate target for monitoring to obtain signal fingerprint information, and positioning the signal fingerprint information through a fingerprint positioning method to obtain a user discrete position of the individual target; and performing road network matching and calibration on the user discrete position to obtain a moving track and accurate positioning information of the individual target so as to track the individual target.

Description

Technical Field

[0001] This invention relates to the field of mobile communication technology, and more specifically, to a method for tracking individual targets based on user behavior. Background Technology

[0002] The entire process of individual target tracking faces the following pain points: it is difficult to completely track the movement trajectory of an individual target; obtaining the location of an individual target is quite difficult; and existing tracking methods have positioning difficulties. Summary of the Invention

[0003] In view of this, the present invention proposes an individual target tracking method based on user behavior to solve the problems existing in the prior art.

[0004] To achieve the above objectives, this invention proposes an individual target tracking method based on user behavior, comprising: Acquire signaling data and construct a travel link database based on the signaling data; Key clues for individual targets are obtained, and these key clues are matched against the travel link database to obtain digital identity information of candidate targets. The candidate target's digital identity information is acquired and monitored to obtain signal fingerprint information. The signal fingerprint information is then located using a fingerprint positioning method to obtain the discrete user location of the individual target. Road network matching and calibration are performed on the discrete locations of the users to obtain the movement trajectory and accurate positioning information of the individual targets for tracking.

[0005] Optionally, the signaling data includes user identifier, timestamp, and base station location information for communication.

[0006] Optionally, the process of constructing the travel link database includes: The signaling data is then cleaned, correlated, and geographically mapped to obtain user location information at different times; the data cleaning includes format verification and invalid data filtering, outlier removal, and data smoothing. Based on the user location information at different times, a time index and database storage are performed to obtain a travel link database.

[0007] Optionally, the process of obtaining the candidate target digital identity information includes: Geographic and time information is extracted from the key clues. The geographic and time information is retrieved from the travel link database to obtain a set of user digital identities. The comprehensive occurrence probability of the user digital identities is calculated to obtain the comprehensive occurrence probability of different user digital identities. The comprehensive occurrence probabilities are sorted, and candidate target digital identity information is obtained based on the sorting results.

[0008] Optionally, the calculation process for the overall occurrence probability includes: Based on the time and location corresponding to the user's digital identity in the user digital identity set, the probability of positioning accuracy, the probability of trajectory coherence, and the probability of time coincidence are calculated based on the time and location. The probability of positioning accuracy, the probability of trajectory coherence, and the probability of time coincidence are weighted and summed to obtain the comprehensive probability of occurrence.

[0009] Optionally, the process of locating the signal fingerprint information using a fingerprint positioning method includes: All user signal fingerprint information and corresponding geographical location information are obtained. The signal fingerprint information is filled and normalized to obtain a feature vector. A random vector is constructed. The feature vector is calculated using the random vector to obtain a composite hash key. The feature vector and the corresponding geographical location information are stored according to the composite hash key to obtain a fingerprint database. The real-time signal fingerprint vector of the candidate target's digital identity information is obtained. The real-time signal fingerprint vector is calculated using a random vector to obtain query features. The query features are then queried in the fingerprint database to obtain candidate vectors. Based on the candidate vectors, the discrete user location of the individual target is obtained.

[0010] Optionally, a road matching method can be used to match and calibrate the user's discrete location with the road network. The road matching method is used to match and calibrate the calculated discrete location points with the road network on the electronic map to obtain the individual target's movement trajectory and accurate positioning information.

[0011] On the other hand, the present invention provides an individual target tracking system based on user behavior for performing the above-described method.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a data-driven individual target tracking system by integrating communication signaling data, video cues, high-precision fingerprint positioning, and intelligent road matching technology, effectively solving core pain points such as difficulty in achieving accurate positioning and trajectory reconstruction. The method associates video cues with signaling data to automatically output a list of high-probability individual targets; it improves positioning accuracy using MR fingerprint positioning and local sensitive hash indexing; it effectively suppresses positioning drift and outputs continuous and reliable motion trajectories by combining a multi-hypothesis particle filter road matching algorithm; and it supports millisecond-level trajectory retrieval through a spatiotemporal composite index, possessing the scalability for city-level data processing. Ultimately, it achieves intelligent management and control of the entire process of individual target tracking, from identification and positioning to real-time tracking and historical backtracking. Attached Figure Description

[0013] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow in an embodiment of the present invention. Detailed Implementation

[0014] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] This embodiment proposes an individual target tracking method based on user behavior, such as... Figure 1 As shown, it includes: S1. Data Acquisition and Preprocessing; Continuously collect and store signaling data from telecommunications operators. This data includes anonymized user identifiers such as International Mobile Subscriber Identity (IMSI), timestamps, and base station location information related to mobile phone communication. The information obtained is only anonymized user identifiers and time-based location information. By cleaning, correlating, and geomapping this data, a dynamic, continuous database of individual travel links in both time and space is built for most anonymized mobile phone users within the region. This transforms individual movements in the physical world into queryable and analyzable digital trajectories, laying the data foundation for subsequent comparison and tracking.

[0016] After collecting the aforementioned signaling data, data cleaning is performed, which mainly includes format verification and invalid data filtering, outlier removal, and data smoothing. The format verification and invalid data filtering process includes: firstly, verifying the completeness of the signaling data records and filtering out records with missing fields (such as missing IMSI or timestamp) or format errors (such as invalid timestamp formats). Simultaneously, data unrelated to personal mobile behavior generated by test terminals, IoT cards, etc., is removed. The outlier removal process includes: based on well-known physical knowledge in the field of communications, setting reasonable threshold ranges (which can be pre-set by relevant professionals) to filter out obviously abnormal measurement values. For example, if the received signal strength (RSRP / RSRP) exceeds the possible reception range of the device (such as above -40dBm or below -140dBm), the measurement report will be considered invalid and discarded. The data smoothing process includes: for signaling data continuously reported by the same user within a short period, records that may have experienced frequent switching of serving cells due to rapid signal fluctuations. Algorithms such as sliding window averaging or Kalman filtering are used to smooth continuously changing parameters such as signal strength in order to suppress instantaneous fluctuations and restore a more realistic signal change trend.

[0017] After data cleaning, relevant data association operations are performed. The cleaned data is still a discrete set of event records, which needs to be transformed into a coherent information flow based on users and behaviors through association. The data association operations include user session association, cross-interface data fusion, and association between the serving cell and neighboring cells. The user session association process involves using an anonymized user identifier (such as IMSI) as the primary key to concatenate signaling events (such as power-on attachment, location update, handover, periodic registration, data service requests, etc.) from scattered signaling data according to their timestamps, forming a continuous session flow for each user. This helps in understanding user behavior within a single network connection cycle. The cross-interface data fusion process involves recognizing that a user's complete behavior may involve signaling from multiple network interfaces (such as Mm interface, Nc interface). The system uses a shared user identifier and association ID (such as Call ID) to stitch together records from different data sources, constructing a comprehensive view of user network activity. The serving cell and neighboring cell association process involves associating the serving cell CGI in each signaling record and the neighboring cell CGI list in the measurement report with an independent base station engineering parameter database. This step is not only to obtain the geographical coordinates of the base station, but also to correlate its key engineering parameters such as antenna azimuth, downtilt angle, and coverage sector number, laying the foundation for building an accurate wireless coverage model in the future.

[0018] After data association, relevant geographic mapping operations are performed. The associated data has known its serving base station. The core task of geographic mapping is to convert the location in cellular network space (CGI) into physical geographic space location (latitude and longitude coordinates). Geographic mapping operations include basic Cell-ID positioning, enhanced triangulation, and spatiotemporal indexing and storage. The specific operations described above are as follows: Basic Cell-ID positioning includes: The most direct method is serving cell centroid positioning. That is, the coverage area of ​​a serving cell is simplified into an area centered on the base station's latitude and longitude, with an estimated coverage radius (the ideal service range distance of the base station) as the circle, and the user's location is initially estimated as the coordinates of that base station. After performing the above basic Cell-ID positioning, to further improve accuracy, enhanced triangulation is performed. The enhanced triangulation process includes: utilizing neighboring cell signal strength information from the measurement report. Through the path loss model in wireless communication, the signal strength (RSRP) of the serving cell and multiple strongest neighboring cells is converted into the approximate distance from the user to each base station. Then, by employing time difference of arrival (TDOA) or power difference of arrival (PDHA) positioning algorithms, a more accurate estimate of the user's location than that of a simple Cell-ID is calculated by solving for the geometric intersection point (which is usually a "confidence region" due to errors).

[0019] After calculating the precise user location estimate as described above, spatiotemporal indexing and storage are performed. This process includes assigning an estimated latitude and longitude coordinate to the data after all the above processing. These coordinates are calculated based on the actual latitude and longitude of the base station and the distance between the user and the base station, using the calculated geometric intersection of the user's estimated location. A spatiotemporal composite index is then built for this data using a spatiotemporal database (such as PostGIS) or a big data platform (such as GeoMesa). During the construction of the spatiotemporal composite index, the user's two-dimensional geographic location, including latitude and longitude, is first encoded into a string using a spatial filling curve (such as Geohash). This string is then combined with a timestamp to form a composite key in the form of "spatial code_timestamp". This key is stored sequentially in the database or big data platform. During a query, the data is quickly located by calculating the encoding prefix corresponding to the spatial range and scanning the time range under these prefixes.

[0020] The aforementioned composite index data can not only be queried by user (IMSI), but also efficiently retrieved by geographical range and time window, thereby enabling core operations such as querying all users who appear in a certain region within a specific time period, providing high-performance support for subsequent analysis.

[0021] After the anonymized user identifier, relevant spatiotemporal information, and subsequent spatiotemporal composite index are determined, the above content is associated under the user identifier to construct a travel link database.

[0022] S2. Individual target selection based on spatiotemporal matching; Obtain relevant key clues, including the geographical location of the individual target in a reasonable public area and the precise time of capture. Based on the camera location and the coverage area of ​​the base station, define one or more virtual areas as focus areas.

[0023] Cameras should be deployed in public areas, such as main roads, transportation hubs, and public squares with high population density.

[0024] Key clues are obtained by identifying content from video footage captured by cameras using large models, or by screening security cameras or cameras in public areas of social resources where the target may appear. Key clues are obtained by identifying a limited number of locations where an individual target may appear through cameras, and by screening multiple pieces of information from these limited locations, one or more digital identities can be identified, and then global tracking can be carried out.

[0025] As a further preferred approach, a large-scale model or deep learning model can be trained on individual target identity photos using a small sample method. The trained large-scale model or deep learning model can then be used to detect and identify each video captured by the camera in order to find key clues. The aforementioned large-scale model and deep learning model techniques for identifying specific individuals are existing knowledge in this field and will not be elaborated further.

[0026] It then calls upon the travel link database built in S1, runs the core algorithm, and identifies mobile phone users who simultaneously appear in all focal areas at a specified time. It calculates the degree of match between each matched user and the spatiotemporal information from the camera, and outputs a list containing the digital identity (IMSI) of potential individual targets and their probability of occurrence. This list is sorted by probability.

[0027] The list generation process is as follows: After obtaining key clues, the geographic coordinates and corresponding precise time points of one or more focus areas are determined. In a pre-processed spatiotemporal trajectory database, all mobile phone users (IMSIs) appearing within the corresponding focus area at each specified time point (allowing an error window of ±1-2 minutes) are queried in parallel. This results in multiple IMSI sets, each representing a user who appeared at a specific location at a specific time point.

[0028] Based on the multiple IMSI sets obtained in the previous step, if there is only one focal area, this step is skipped. If there are multiple focal areas (e.g., locations A and B), the algorithm performs a set intersection operation. This then identifies a list of IMSIs that appear in all focal areas simultaneously, containing several possible candidate IMSIs.

[0029] For each candidate IMSI, a comprehensive occurrence probability P is calculated. This comprehensive probability includes the probability of location accuracy, the probability of trajectory consistency, and the probability of time coincidence. These probabilities are obtained through a weighted sum. The calculation details for each probability are as follows: Location accuracy probability: Based on the accuracy of the location technology (such as Cell-ID or MR), the reliability of the location point is evaluated. Higher accuracy results in a higher location accuracy probability Ploc. The formula for calculating the location accuracy probability Ploc is Ploc = exp(-d 2 / 2*R 2 max ), where d represents the estimated actual Euclidean distance between the user's location and the target point (such as a camera), and R max Let denot represent the typical error radius of the current positioning technology (e.g., 500 meters for base station positioning and 150 meters for fingerprint positioning), and exp represent an exponential function with the natural constant e as the base.

[0030] Trajectory continuity probability: This assesses the rationality of a user's path from one appearing area to the next. For example, it calculates the straight-line distance between two points and the actual time taken to determine if the movement speed is within a feasible range (e.g., walking, driving). The more rational the path, the higher the probability of trajectory continuity. When V1 is greater than V... max At that time, the probability of trajectory continuity is Ptraj = exp(-(V1-V)). max ) 2 / 2*b 2 ), where V1 represents the observed movement speed, which is the ratio of the spatial distance between two consecutive clue points to the time interval, V max This represents the preset reasonable maximum moving speed (e.g., 80 km / h). b represents the speed tolerance parameter, controlling the rate of probability decay when the observed speed exceeds the maximum speed. When V1 is less than or equal to V... max When Ptraj=1.

[0031] Time coincidence probability: The accuracy with which the user appears within the camera's time error window. The closer to the center time point, the higher the time coincidence probability. Ptime=max(0,1-|Δt| / Tmax). Δt represents the time difference between the user's appearance time and the camera's capture time. Tmax represents the maximum allowable time error, i.e., a reasonably set time window.

[0032] The candidate IMSIs are sorted in descending order according to the calculated overall probability to generate a list, thereby determining the different user digital identities and their corresponding probabilities.

[0033] This enables a shift from searching for people in massive video feeds to screening people based on precise data, transforming the target from hard-to-identify faces into traceable, anonymous digital identities (IMSIs), thus solving the pain points of low video coverage and heavy manual screening workload.

[0034] S3. Trajectory presentation based on high-precision positioning; From the list output by S2, identify one or more high-probability targets (their IMSIs). Real-time location and trajectory reconstruction are required: for the high-probability target IMSIs, real-time signaling data stream monitoring is initiated. MR (Measurement Report) fingerprint positioning technology is employed, analyzing parameters such as signal strength and quality between the mobile phone and multiple base stations, combined with an existing signal fingerprint database, to calculate a more accurate user location than base station positioning alone.

[0035] The fingerprint positioning technology described above is implemented through the following scheme, which includes: Phase 1: Offline Database Construction (Fingerprint Collection and LSH Index Building) This phase involves creating a fingerprint database and building an efficient LSH index structure. Specifically, geographic coordinates and their corresponding signal fingerprint vectors V=[RSRP_A, RSRP_B,...] are collected through drive testing or crowdsourcing. RSRP_A and RSRP_B refer to the reference signal received power from base station B as measured by the mobile phone. To address the signal strength sparsity problem (i.e., the mobile phone will not measure all base stations), the signal fingerprint vector is padded and normalized to form a fixed-dimensional, dense feature vector. For cosine similarity, a corresponding family of LSH hash functions is selected, such as random hyperplane hashing. L hash tables are created, each consisting of K randomly generated hash functions.

[0036] In the hash table generation process, a fixed-length random vector *r* is first randomly generated as the normal vector of the hyperplane, representing the signal fingerprint vector. Each dimension (component) of this random vector *r* is independently randomly drawn from a standard normal distribution (mean 0, variance 1). This ensures that the generated hyperplane is completely random and uniformly distributed in space. Repeating this random vector generation method independently *K* times yields *K* distinct, independently generated D-dimensional random vectors: [r1, r2, r3, ..., r...]. K These K random vectors together form a group of K hash functions in a hash table.

[0037] For each fingerprint vector V in the database, it is fed into one of the L hash tables. In each hash table, the outputs of K hash functions are combined into a "composite hash key," which determines a bucket. The vector V and its corresponding geographic coordinates are stored in its corresponding L buckets.

[0038] After generating these K hash functions, for a given data vector V, the first random vector r1 is used to perform a dot product with V, and the result is output as 1 or 0 depending on its sign, yielding the first bit. The second random vector r2 is used to perform a dot product with V, and the result is output as 1 or 0 depending on its sign, yielding the second bit. This process continues with the Kth random vector r1... K The dot product with V is used to obtain the Kth bit, depending on its sign. Finally, these K bits are combined to form a K-bit binary string (e.g., 1011...0). This K-bit binary string is the composite hash key of the data vector V in this hash table, determining which bucket V will be placed in. Specifically, the composite hash key represents the address label of the bucket in the hash table, where the bucket represents a storage space, and the address label of the storage space is the composite hash key. When the fingerprint vector of the above fingerprint signal matches the address label of the storage space, the fingerprint vector is placed in the storage space with the same address label as the composite hash key. Similarly, the vector and its corresponding geographic coordinates are stored in the storage space corresponding to the bucket.

[0039] The second stage: Online location (real-time hashing and candidate set refinement), specifically includes: obtaining the real-time MR report of the target mobile phone and constructing its real-time signal fingerprint vector V_query using the exact same process as the offline stage. The query vector V_query is processed using the same L sets of hash functions to obtain L composite hash keys. Based on these L keys, L corresponding buckets are retrieved from L hash tables. Truly similar fingerprint vectors will appear with high probability in the union of these L buckets. This process avoids comparison with the entire database. All fingerprint vectors in the above L buckets are merged to form a candidate set much smaller than the entire database. Within this small candidate set, a precise similarity algorithm (such as cosine similarity or Euclidean distance) is used to calculate the distance between V_query and each candidate vector. The Top-M most similar fingerprints (M can be very small, such as 10) are found from the candidate set, and their corresponding geographic coordinates are weighted and averaged, with the weight proportional to the similarity, ultimately yielding the estimated location of the mobile phone.

[0040] In this step, the search time is reduced from being proportional to the database size to being proportional to the bucket size, making it particularly suitable for fingerprint databases with tens of millions or even larger scales. By adjusting the parameters L (number of hash tables) and K (number of hash functions per table), a flexible trade-off can be struck between location accuracy and query speed.

[0041] Following fingerprint positioning technology, a road matching algorithm is introduced to match and calibrate the calculated discrete location points with the road network on the electronic map, so as to smoothly and accurately lock the user's movement trajectory on the actual road and eliminate the positioning drift phenomenon.

[0042] The video footage taken in reasonable public areas provides limited public video information, making it impossible to trace the movements of the people involved. Therefore, for the unavailable trajectory information, hypothetical road trajectories are matched using subsequent footage.

[0043] The implementation process of the road matching algorithm described above is as follows: 1) Initialize the multipath hypothesis: After receiving the first location point of the trajectory, instead of directly matching it to the nearest road, a large number (e.g., thousands) of "particles" are generated within a reasonable range around it. Each particle carries complete "identity" information, including: its current location on which specific road, its historical paths, its current motion state (such as speed and direction), and a "weight" representing its credibility.

[0044] 2) Intelligent prediction based on road network topology: When a new location point arrives, each particle makes predictions based on its current position and movement state along the actual road network structure. Each particle strictly adheres to road connectivity. Particles can only move to downstream road segments directly connected to the current road. At intersections, a particle splits into multiple particles, each entering different possible branches, thus naturally maintaining multiple hypothesis tracking at decision points.

[0045] 3) Weight Update for Multi-Dimensional Evidence Fusion: Each particle calculates its fit with the newly arrived location point and updates its own weight accordingly. This process integrates multiple pieces of evidence: Spatial proximity: The closer the particle's position is to its original location in a straight line, the higher the score. Orientation consistency: The smaller the angle between the direction of the particle's path and its direction of movement, the higher the score. Path connectivity: Whether the particle's current predicted position can be naturally reached from its historical position, i.e., whether the path is topologically connected. Motion smoothness: Whether the particle's calculated velocity and acceleration are within a reasonable physical range.

[0046] Among them, the distance score = exp(-0.5 * (particle position - observation point distance) * 2 / standard deviation of positioning error); the direction score = cos(particle heading angle - road direction angle); the topology score is determined by whether the particle's historical path is connected, where connected is 1 and disconnected is 0; the motion score = exp(-|particle predicted velocity - actual calculated velocity| / preset velocity), and the update weight is the product of the above four scores. The new weight is the product of the weight in the previous step and the updated weight, and the weight is updated accordingly.

[0047] 4) Adaptive resampling to maintain efficiency: The system periodically checks the overall quality of all particles. If a large number of particles have very low weights, it indicates that many assumptions are no longer valid, and the system initiates a "resampling" process. Simultaneously, particle hypotheses with extremely low weights (below a certain threshold) are removed. Excellent particle hypotheses with high weights (above a certain threshold) are replicated to fill the gaps left by the eliminated particles. Furthermore, a small subset of optimal particles are directly retained, unaffected by resampling, preventing the accidental loss of the best hypotheses. In this way, computational resources are always focused on the most likely path hypotheses.

[0048] 5) Multi-path confidence output: The system ultimately outputs not a single path, but a "list of path hypotheses" sorted by confidence (weight). Through clustering analysis, such as k-means clustering, particles with similar paths are grouped into one class, each representing a possible trajectory interpretation. The path with the highest confidence within each cluster is then displayed as a high-precision trajectory. Simultaneously, the final endpoint positioning information from this high-precision trajectory is shown to represent the location of individual targets on the final path, thus eliminating drift.

[0049] In this step, through multi-evidence fusion and continuous multi-hypothesis verification, correct judgments can be made based on the overall trend, even when GPS signal drift is severe. It is also insensitive to individual anomalous location points, as a few "misleading" particles are overwhelmed by a large number of "correct" particles. Furthermore, it provides multipath hypotheses and their confidence levels.

[0050] The processed, high-precision trajectory data is pushed out in real time and displayed on an electronic map, forming an intuitive movement path. Simultaneously, the target's historical trajectory can be queried and replayed to analyze its behavioral patterns.

[0051] This step enables continuous, virtual, and precise monitoring of the target individual, addressing the pain point of difficulty in obtaining location information.

[0052] On the other hand, the present invention also provides an individual target tracking system based on user behavior, comprising: The first module is used for data acquisition and preprocessing; The second module is used for individual target selection based on spatiotemporal matching; The third module is used for trajectory presentation based on high-precision positioning.

[0053] The system provided in this invention corresponds to the method described above, and will not be repeated here.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for tracking individual goals based on user behavior, characterized in that, include: Acquire signaling data and construct a travel link database based on the signaling data; Key clues for individual targets are obtained, and these key clues are matched against the travel link database to obtain digital identity information of candidate targets. The candidate target's digital identity information is acquired and monitored to obtain signal fingerprint information. The signal fingerprint information is then located using a fingerprint positioning method to obtain the discrete user location of the individual target. Road network matching and calibration are performed on the discrete locations of the users to obtain the movement trajectory and accurate positioning information of the individual targets for tracking.

2. The method according to claim 1, characterized in that, The signaling data includes user identifier, timestamp, and base station location information for communication.

3. The method according to claim 1, characterized in that, The process of constructing the travel link database includes: The signaling data is then cleaned, correlated, and geographically mapped to obtain user location information at different times; the data cleaning includes format verification and invalid data filtering, outlier removal, and data smoothing. Based on the user location information at different times, a time index and database storage are performed to obtain a travel link database.

4. The method according to claim 1, characterized in that, The process of obtaining the candidate target's digital identity information includes: Geographic and time information is extracted from the key clues. The geographic and time information is retrieved from the travel link database to obtain a set of user digital identities. The comprehensive occurrence probability of the user digital identities is calculated to obtain the comprehensive occurrence probability of different user digital identities. The comprehensive occurrence probabilities are sorted, and candidate target digital identity information is obtained based on the sorting results.

5. The method according to claim 4, characterized in that, The calculation process for the overall occurrence probability includes: Based on the time and location corresponding to the user's digital identity in the user digital identity set, the probability of positioning accuracy, the probability of trajectory coherence, and the probability of time coincidence are calculated based on the time and location. The probability of positioning accuracy, the probability of trajectory coherence, and the probability of time coincidence are weighted and summed to obtain the comprehensive probability of occurrence.

6. The method according to claim 1, characterized in that, The process of locating signal fingerprint information using fingerprint positioning methods includes: All user signal fingerprint information and corresponding geographical location information are obtained. The signal fingerprint information is filled and normalized to obtain a feature vector. A random vector is constructed. The feature vector is calculated using the random vector to obtain a composite hash key. The feature vector and the corresponding geographical location information are stored according to the composite hash key to obtain a fingerprint database. The real-time signal fingerprint vector of the candidate target's digital identity information is obtained. The real-time signal fingerprint vector is calculated using a random vector to obtain query features. The query features are then queried in the fingerprint database to obtain candidate vectors. Based on the candidate vectors, the discrete user location of the individual target is obtained.

7. The method according to claim 1, characterized in that, The road matching method is used to match and calibrate the road network of the user's discrete location. The road matching method is used to match and calibrate the calculated discrete location points with the road network on the electronic map to obtain the movement trajectory and accurate positioning information of the individual target.

8. An individual target tracking system based on user behavior, characterized in that, Used to perform the method described in any one of claims 1-7.

Images