A data collection method and system based on dwell depth

Abstract

The application relates to a data collection method and system based on a stay depth, and relates to the field of data collection, which comprises the following steps: collecting user positions and user information; obtaining a stay duration according to the user positions and a preset reference point; calling detection time, target features and actual output information from the user information; marking an actual position based on the detection time and the user positions; obtaining an access quantity based on the actual position and the reference point; obtaining final output information by combining the stay duration, the access quantity, the reference point, the actual output information and the target features, and uploading a report. The application has the effect of improving data collection accuracy.

Description

Technical Field

[0001] This invention relates to the field of data acquisition technology, and in particular to a data acquisition method and system based on dwell depth. Background Technology

[0002] Data collection is a core element of offline store scenarios and a key factor affecting operational efficiency, marketing accuracy, business format adaptability, and commercial revenue conversion.

[0003] Existing data collection technologies rely on manual methods to screen and statistically analyze core data such as purchase frequency, average transaction value, and number of visits through store back-end POS systems, consumption ledgers, and member registration terminals. This data is then manually integrated through multiple manual methods, including offline questionnaires, one-on-one interviews to collect user needs and preferences, member information registration, and on-site observations.

[0004] In the data collection process, manually collecting user data is cumbersome, and manual data integration can lead to discrepancies in recording and statistics, reducing the efficiency and accuracy of the final output information analysis. Summary of the Invention

[0005] To improve the accuracy of data acquisition based on dwell depth, this invention provides a data acquisition method and system based on dwell depth.

[0006] In a first aspect, the present invention provides a data acquisition method based on dwell depth, employing the following technical solution: A data acquisition method based on dwell depth includes: Collect user location and user information; The dwell time is determined based on the user's location and a preset reference point; Retrieve detection time, target features, and actual output information from user information; The actual location is marked based on the detection time and the user's location; The number of visits was obtained based on the actual location and reference points; The final output information is obtained by combining the dwell time, number of visits, reference points, actual output information and target characteristics, and then the report is uploaded.

[0007] By adopting the above technical solution, the user location and user information are collected to obtain dwell time, number of visits, detection time, target characteristics and actual output information. Based on the above data, the final output information is determined and a report is generated and uploaded. The data is automatically collected and integrated to obtain the final output information, reducing the need for manual interviews, observation and manual recording of survey data, so as to improve the efficiency and accuracy of data collection and analysis of dwell depth.

[0008] Optionally, methods for correcting the final output information include: The remaining location is obtained based on the user's location and the actual location; The remaining path and remaining dwell time are obtained by comparing the remaining location with the preset reference range; The number of marked references is obtained based on the remaining path and the reference range; The detection level is determined by comparing the remaining dwell time with the preset baseline dwell time range. The range of marked categories is obtained by combining the remaining locations, detection levels, and preset reference information; The tag features are obtained based on the tag category range and reference information; The final output information is obtained by combining the detection level, the number of reference markers, and the features of the reference points and markers, and then uploaded.

[0009] By adopting the above technical solution, the final output information of the remaining location can be obtained by analyzing the user's location and actual location, path and stay data. This can improve the accuracy of the data analysis of stay depth even if the customer has not made a purchase.

[0010] Optionally, the correction methods for the remaining position include: Retrieve baseline path features from reference information; The selected path is determined based on the reference range and the remaining path. Based on the baseline path features and the selected path, the overlapping path and the final path are derived. The detection and placement image is obtained by collecting image detection information based on the final path. The actual baseline path is obtained by detecting the placement of images and the final path; The detection positions are obtained by comparing the actual baseline path with the overlapping path, and the detection positions are removed from the remaining positions to update the remaining positions.

[0011] By adopting the above technical solution, the remaining customer paths are filtered and compared to obtain overlapping paths and the final path by retrieving the baseline path features in the store. Combined with image detection information, the actual employee operation path is identified. After removing the detection positions corresponding to the employee paths, the remaining positions are updated, reducing the interference of non-customer trajectories on the final output information analysis of the remaining positions and improving the accuracy of the final output information analysis.

[0012] Optionally, the methods for correcting the final output information also include: Collect detection images of the user's location and build a detection model based on the detection images and preset human features; The detection interval is obtained based on the user's location; The aggregation range is obtained by comparing the detection interval with the preset row spacing; The aggregation range is updated at a preset detection time, and the detection range is obtained by comparing the aggregation range before and after the update. The detection category is obtained by using the detection model and the detection range; The final output information is updated based on the detection category and reference point.

[0013] By adopting the above technical solution, a detection model is established by collecting detection images of user locations. The customer gathering range is determined by combining the detection distance and the distance between people in the same row. After verification by the detection time, the detection range is determined and its detection category is identified. Finally, the final output information is updated based on the detection category. This can accurately identify the main consumers in the crowd and improve the accuracy of the data analysis of dwell depth.

[0014] Optionally, methods for updating the final output information include: The reference points are obtained based on the actual location and the remaining path. The reference category is obtained based on the reference information and the reference points visited. The target category is obtained by comparing the detected category with the benchmark reference type; The walking speed is obtained by using the remaining position and the reference range; Retrieve the reference type from the reference category; The marker's movement speed is obtained based on the detection category; The detection reference point is obtained by comparing the walking speed with the marker speed and combining the reference range; The final output information is updated based on the target category and the detection reference point.

[0015] By adopting the above technical solution, matching customer detection categories with store benchmark reference types, and combining walking speed characteristics to lock customer target reference points, the final output information is updated to identify potential consumption tendencies and further improve the accuracy of the final output information.

[0016] Optionally, the verification methods for the detection category include: Retrieve the gaze direction and head orientation from the detection model; The field of view is determined by the direction of the gaze and the orientation of the head. The target reference point is obtained based on the field of view and the reference points it passes through; The characteristics of the consumer subject are obtained through the target reference point; The detection category is determined by combining the characteristics of the consumer group with the detection model.

[0017] By adopting the above technical solution, the types of stores where customers have made historical purchases are statistically analyzed, the actual output cycle and the baseline output cycle are calculated, the marked output cycle is determined after verification, abnormal output cycles are identified and the marked features are updated accordingly, the interference of abnormal consumption behavior is filtered out, and the accuracy of obtaining the target features of customer preferences is improved.

[0018] Optionally, methods for updating labeled features include: Retrieve historical reference types from user information; The actual output cycle and the baseline output cycle are obtained based on the historical reference type; The marked output cycle is obtained based on the consistency between each actual output cycle and the reference output cycle; The abnormal output period is obtained based on the marked output period and the actual output period; Update the labeling features based on the abnormal output cycle and historical reference type.

[0019] By adopting the above technical solution, other detection data authorized by customers are collected, other reference ranges of customers are analyzed, and compared with the target features of offline stores. The label features are updated according to the matching results, and the correlation analysis of online and offline data is opened up to improve the completeness of data analysis on dwell depth.

[0020] Optionally, methods for updating labeled features also include: Collect other detection data using user information; Other text content and other operation data are obtained based on other detection data; Other reference ranges are obtained based on other text content and other operational data; The target feature range is obtained based on reference information and target features; By comparing the target feature range with other reference ranges, if the target feature range is within other reference ranges, the marking range is obtained by comparing the target feature range with other reference ranges. Obtain the target features based on the labeling range and update the labeling features; If the target feature range is not within other reference ranges, update the labeled features based on other reference ranges.

[0021] By adopting the above technical solution, the inferred target characteristics of customers can be sorted out from other detection data, and a consistency comparison can be made with offline consumption data. Based on the results, consumption types can be distinguished and the labeling characteristics can be updated, reducing misjudgments caused by abnormal consumption behaviors such as purchasing agents and improving the accuracy of dwell depth data.

[0022] Optionally, methods for updating labeled features also include: Target features are inferred based on other detection data; The consistency between the inferred target characteristics, historical reference types, and target characteristics is determined by comparing them. If they match, retrieve the total output information for the abnormal output period from the user information. Update the labeled features based on the total output information; If there is a discrepancy, the abnormal target features are obtained based on the inferred target features, historical reference types, and target features. The labeling features are updated based on the characteristics of the abnormal target and the abnormal output cycle.

[0023] By adopting the above technical solution, the inferred target features obtained based on platform data are compared with the customer's historical reference type and actual consumption target features. Based on the differences in the comparison results, the label features are updated by combining the total output information or abnormal target features within the abnormal output period. This can effectively distinguish between the customer's real consumption and abnormal consumption behavior, and reduce the interference of abnormal consumption data on the update of label features.

[0024] Secondly, this application provides a data acquisition system based on dwell depth, employing the following technical solution: A data acquisition system based on dwell depth includes: The acquisition module is used to obtain user location and user information; A memory used to store a program for a data acquisition method based on dwell depth; The processor is used to load and execute programs stored in memory.

[0025] In summary, this application includes at least one of the following beneficial technical effects: 1. By collecting user location and user information, the system obtains dwell time, number of visits, and consumption data (detection time, target characteristics, and actual output information). Based on the dwell time, number of visits, and consumption data, the system determines the final output information, generates a report, and uploads it. The system automatically collects and integrates data to obtain the final output information, reducing the need for manual interviews, on-site observations, and manual recording of survey data, thereby improving the efficiency and accuracy of dwell depth data analysis. 2. By retrieving the baseline path features within the store, the remaining customer paths are filtered and compared to obtain overlapping paths and the final path. Combined with image detection information, the actual employee work paths are identified. After removing the detection positions corresponding to the employee paths, the remaining positions are updated, reducing the interference of non-customer trajectories on the final output information analysis of the remaining positions and improving the accuracy of the final output information analysis. 3. Statistically analyze the store types of customers' historical consumption, calculate the actual output cycle and the baseline output cycle, determine the marked output cycle after verification, identify abnormal output cycles and update the marked features accordingly, filter out the interference of abnormal consumption behavior, and improve the accuracy of obtaining the target features of customer preferences. Attached Figure Description

[0026] Figure 1 This is a flowchart of a data acquisition method based on dwell depth according to an embodiment of the present invention; Figure 2 This is a flowchart of the method for correcting the final output information in an embodiment of the present invention. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0028] Reference Figure 1 This application discloses a data acquisition method based on dwell depth, comprising the following steps: S10: Collect user location and user information.

[0029] User location refers to the specific coordinates of a person within the shopping mall. The specific coordinates of a person within the shopping mall, obtained through radar monitoring, are used as the user's location.

[0030] User information refers to information generated when customers make purchases in a shopping mall, including the detection time, the store visited, and the amount spent.

[0031] The membership account system retrieves data or content from the consumer's membership account as user information. The membership account stores the consumer's publicly available user information.

[0032] S11: Calculate the dwell time based on the user's location and a preset reference point.

[0033] The reference points are the characteristics of all shops in the shopping mall, including their number, category, and area, as set by the technical staff.

[0034] Dwell time refers to the length of time people stay in a store within a shopping mall.

[0035] The reference point retrieval range is set. When the user's location is within the range, the timer starts and stops when the user's location disappears. The timer result corresponding to each store is used as the dwell time.

[0036] The dwell time is determined by the time interval during which the user's location is within the store's area.

[0037] S12: Retrieve detection time, target features, and actual output information from user information.

[0038] The inspection time refers to the point in time when the customer checks out.

[0039] Target characteristics refer to the category, color, shape, etc. of a specific product when a customer checks out for it.

[0040] The actual output information refers to the total price of the corresponding goods at the time of checkout.

[0041] The detection time, target features, and actual output information are retrieved from the user information.

[0042] S13: Mark the actual location based on the detection time and user location.

[0043] Actual location refers to the location of a user whose identity is known to staff within the shopping mall after a purchase.

[0044] The actual location is selected from the user locations at the counter during the detection time point.

[0045] S14: The number of visits is obtained based on the actual location and reference point.

[0046] The number of visits refers to the number of stores that customers who have made a purchase enter during their browsing process.

[0047] The consumer's original walking path is determined by their actual location. The total number of stores visited is calculated based on the instances where the path is located at a reference point.

[0048] S15: Combine dwell time, number of visits, reference points, actual output information and target characteristics to obtain the final output information and upload the report.

[0049] The final output information refers to the content of collecting and integrating data related to the store, including information on browsers, superficial consumers, and deep consumers.

[0050] By comparing the number of visits with the number of reference points, we can determine the actual percentage of stores visited by customers, thus calculating the browsing depth. Customers' actual output information, dwell time, and purchase goals contribute to the conversion rate during the browsing process. The browsing depth and conversion rate are then input into an integrated hierarchical model to obtain the final output information, which is then uploaded to the operator as a report.

[0051] The integrated hierarchical model is trained by pre-inputting data such as dwell time, number of visits, reference points, actual output information and target features from the operator, and then directly outputs the final output information.

[0052] For example, the browsing depth can be obtained by the percentage of actual stores visited; the conversion rate can be obtained by the actual information output, dwell time, and target characteristics.

[0053] Visitors: Browsing depth rate ≤15%, consumption conversion rate ≤5%.

[0054] Superficial consumers: browsing depth rate 15%~40%, consumption conversion rate 5%~30%.

[0055] Deep consumers: browsing depth rate ≥40%, consumption conversion rate ≥30%.

[0056] Reference Figure 2 The methods for correcting the final output information include: S16: Obtain the remaining location based on the user's location and the actual location.

[0057] The remaining location refers to the location within the shopping mall where people do not make a purchase and whose identity is unknown.

[0058] The remaining locations (locations of non-consumers) are obtained by removing the actual locations from all customer location data in S13, and are used to analyze the browsing paths of non-consumers.

[0059] S17: Obtain the remaining path and remaining stay time by comparing the remaining location with the preset reference range.

[0060] The remaining path refers to the trajectory of a customer moving within the mall without making a purchase.

[0061] Remaining dwell time refers to the length of time a customer stays within each reference range without making a purchase.

[0062] The remaining dwell time is obtained by forming the remaining paths of the location points within the reference range according to the time series, and by calculating the dwell time of the location points or paths in each store.

[0063] S18: Calculate the number of marked references based on the remaining path and the reference range.

[0064] The reference quantity refers to the number of stores a customer enters without making a purchase.

[0065] The total number of stores with a path range will be used as the reference number for marking.

[0066] S19: The detection level is obtained by comparing the remaining dwell time with the preset baseline dwell time range.

[0067] The baseline duration range is a time standard interval pre-set by technicians to classify the remaining dwell time into different levels. For example, the duration range is 0-30 seconds, 31-120 seconds, and more than 120 seconds, with each range corresponding to a detection level of low, medium, or high.

[0068] The detection level refers to the degree of interest a customer has in a store they linger in, even before making a purchase.

[0069] The customer's detection level in each store is determined by comparing the remaining stay time with the baseline time range.

[0070] S20: Combine the remaining location, detection level, and preset reference information to obtain the range of marking categories.

[0071] The reference information is pre-set by technicians and includes information such as the store's category, floor area, merchandise placement, counter location, and replenishment warehouse location.

[0072] The category range refers to the range of store categories that customers show interest in even before making a purchase.

[0073] By filtering out stores with high detection levels from the remaining locations as stores of interest (stores where customers spend a long time), and combining this with reference information, the categories of stores of interest (clothing, food and beverage, digital products, beauty products, etc.) are extracted to form a range of marked categories.

[0074] S21: Obtain the tag features based on the tag category range and reference information.

[0075] Marking features refer to the category, color, shape, etc. of the goods at the target reference point corresponding to the customer in the unconsumed state.

[0076] The characteristics of goods within the marked category range (color, category, shape, brand, etc.) are retrieved from the reference information as marked features.

[0077] S22: Combine the detection level, the number of reference markers, and the features of the reference points and markers to obtain the final output information and upload it.

[0078] The percentage of marked stores is obtained by comparing the number of marked references with the number of reference points (the higher the percentage of marked stores, the deeper the browsing, and vice versa). The detection level and marking features are combined to obtain a potential consumption conversion rate (the lower the potential consumption conversion rate, the lower the customer value, and the higher the potential consumption conversion rate, the higher the customer value). The percentage of marked stores and the consumption conversion rate are uploaded to the integrated hierarchical model to obtain the final output information of the remaining positions and then uploaded.

[0079] The methods for correcting the remaining position include: S23: Retrieve baseline path features from reference information.

[0080] The baseline path features refer to the characteristics of the routes that simulate the walking routes of employees during operations such as checkout at the counter, retrieving goods from the warehouse, replenishing goods at designated locations, and organizing shelves.

[0081] By extracting the walking routes of store staff at counter locations, shelf layouts, and warehouse locations from the specification data of reference points, these routes are used as baseline path features for subsequent comparative analysis with actual customer paths.

[0082] S24: Obtain the selected path based on the reference range and the remaining path.

[0083] The selected path refers to the remaining paths within the reference range.

[0084] By extracting paths within the reference range from the remaining paths as selected paths, the specific walking trajectories of customers within the store can be analyzed.

[0085] S25: Based on the baseline path features and the selected path, the overlapping path and the final path are derived.

[0086] Overlapping paths refer to selected paths that overlap with the features of the baseline path.

[0087] The final path refers to the remaining walking trajectory after removing overlapping paths from the selected paths.

[0088] By comparing the baseline path features with the selected path, the overlapping segments (such as the paths taken by both parties to pick up goods from the warehouse and replenish the shelves) are identified as overlapping segments, and the parts of the selected path that do not overlap with the overlapping paths are taken as the final path.

[0089] S26: Obtain the detection placement image based on the image detection information acquired according to the final path.

[0090] The image display refers to the image information showing the product display along the final path within the store.

[0091] Images of merchandise placement on shelves and display cases along the final path are retrieved from cameras along the path and used as images for placement detection.

[0092] S27: Obtain the actual baseline path by detecting the placement of images and the final path.

[0093] The actual baseline path refers to the path identified in the detection and placement images where there are traces of employee operations (such as hand movements during replenishment, tool placement, and shelf adjustment marks).

[0094] Image analysis is performed by detecting the placement of images. For example, by cropping product images before and after a person passes by, the neatness of the product arrangement and whether the products are facing forward can be compared to determine whether the person is a worker. The path of the person is then marked to obtain the actual path of the employee performing the work, which serves as the actual baseline path.

[0095] S28: Obtain the detection position by comparing the actual reference path with the overlapping path, and remove the detection position from the remaining positions to update the remaining positions.

[0096] The detection location refers to the position of the staff among the remaining locations.

[0097] The remaining positions are updated by using the remaining positions corresponding to the actual baseline path and the overlapping path as detection positions, and removing the detection positions from the remaining positions.

[0098] The methods for correcting the final output information also include: S29: Collect detection images of the user's location and establish a detection model based on the detection images and preset human features.

[0099] The detected image refers to the image of the customer's location captured by the camera.

[0100] A detection model is a feature model established based on image detection results, used to identify and distinguish different individual customers.

[0101] Character features are elements such as the human body shape outline that are pre-defined for technicians.

[0102] The image acquisition device acquires a detection image corresponding to the user's location. Human features are used to detect and extract features from the image, and the extracted information is integrated to create a 3D model for the customer, which serves as the detection model. In this embodiment, the image acquisition device only captures and identifies the lines and contours of the human figure.

[0103] S30: Detection spacing is obtained based on the user's location.

[0104] The detection interval refers to the distance between any two customers.

[0105] The detection interval is calculated by using the straight-line distance between any two user locations.

[0106] S31: The aggregation range is obtained by comparing the detection interval with the preset row spacing.

[0107] The peer spacing is a distance threshold preset by technicians to determine whether multiple customers belong to the same peer group.

[0108] The cluster area refers to the spatial area formed by multiple customers who meet the spacing requirements.

[0109] By comparing the detection distance with the peer distance, customers whose distance is less than or equal to the peer distance threshold are selected, and adjacent customers are aggregated into the same group, and the spatial range of the group is defined as the aggregation range.

[0110] S32: Update the aggregation range with a preset detection time, and compare the aggregation range before and after the update to obtain the detection range.

[0111] The detection time is a preset time value by the technicians to verify whether the person is traveling with another person.

[0112] The detection range refers to the cluster range that has been verified and meets the peer criteria.

[0113] By repeatedly executing step S31 during the detection time, a new clustering range is obtained. The clustering ranges before and after the update are compared for consistency, and the clustering range that meets the consistency condition is determined as the detection range.

[0114] S33: The detection category is obtained by using the detection model and the detection range.

[0115] The detection category refers to the type of group that is divided into peer groups, such as family groups, friend groups, couple groups, business groups, etc.

[0116] The detection model established by S29 is associated with the detection range obtained by S32. The human characteristics (such as height, gender, and interaction posture) of each customer within the detection range are extracted and input into the preset classification model to identify the relationship type between group members and obtain the detection category.

[0117] The classification model is trained by staff who pre-extract human contour features such as height, gender, and interactive posture of each customer, and directly identify the relationship type of group members and output the detection category. The classification model is set in advance by those skilled in the art and will not be described in detail here.

[0118] S34: Update the final output information based on the detection category and reference point.

[0119] The detection category and reference point are analyzed to obtain new final output information.

[0120] Methods for updating the final output information include: S35: Obtain reference points based on the actual location and remaining path.

[0121] Reference points refer to the collection of stores a customer actually enters during their actual walk, regardless of whether they make a purchase.

[0122] Each store in the number of visits is defined as a reference point by comparing it with each store in the number of marked references in S18.

[0123] S36: The reference category is obtained based on the reference information and the reference point.

[0124] Reference category refers to the classification of business type or product category to which the reference point belongs.

[0125] By combining the reference category information stored in S23 with the reference points, the category corresponding to each reference point (such as clothing, catering, digital products, beauty products, etc.) is extracted as the reference category.

[0126] S37: Retrieve the baseline reference type from the reference category.

[0127] The benchmark reference type refers to the target customer group type (such as family customers, young customers, business customers, couples, etc.) that is preset in the reference information and is mainly targeted by a certain type of store.

[0128] The target customer group type is retrieved from the reference information as the benchmark reference type.

[0129] S38: The target category is obtained by comparing the detection category with the benchmark reference type.

[0130] The target category refers to the set of reference categories that match the customer's detection category with the store's baseline reference type.

[0131] By comparing the detection category with the benchmark reference type corresponding to the reference category, the category whose detection category is consistent with the benchmark reference type is selected as the target category.

[0132] S39: Obtain the walking speed by using the remaining position and the reference range.

[0133] Walking speed refers to the speed characteristics of a customer walking within a reference range when not making a purchase.

[0134] By combining the remaining locations with the store's floor area, the customer's location points within each store are extracted, and the distance change between adjacent locations is calculated to obtain the customer's walking speed within the store.

[0135] S40: Obtain the marker movement speed based on the detection category.

[0136] Marker movement speed refers to the general speed at which a group moves within a target reference point.

[0137] The movement speed threshold corresponding to the detection category is retrieved from the speed model preset by the technicians and used as the marker movement speed.

[0138] A velocity model is a model that is trained by technicians by pre-inputting detection categories and group movement characteristics, and outputs a threshold signal for marked movement speed. The velocity model is pre-set by the staff.

[0139] S41: The detection reference point is obtained by comparing the walking speed with the marker speed and combining the reference range.

[0140] The detection reference point refers to a store where the customer's walking speed matches the speed characteristics marked by the detection category.

[0141] The walking speed is compared with the marked walking speed to determine whether the customer's walking speed in each store is within the normal deviation range of the marked walking speed (e.g., within ±3% of the range of interest). Based on this reference range, stores that meet the speed characteristics are selected as reference points for testing.

[0142] S42: Update the final output information based on the target category and the detection reference point.

[0143] By analyzing the target category, the number and percentage of stores matching the detected category among the reference points visited by customers are counted to obtain the category matching degree (the higher the category matching degree, the higher the customer's conversion rate, and vice versa). Then, by analyzing the detection reference points and their corresponding reference categories, the customer's interests and preferences under the detection reference points are analyzed. Based on S15 and S22, the interests and preferences and category matching degree are input into the integrated hierarchical model to update the final output information.

[0144] The methods for verifying the detection category include: S43: Retrieve the gaze direction and head orientation from the detection model.

[0145] The direction of gaze refers to the direction in which the customer's eyes are focused.

[0146] Head orientation refers to the angle at which the customer's head is facing.

[0147] By extracting head posture and gaze data for each customer from the detection model, we can obtain head orientation angles (such as pitch and yaw angles) and gaze directions, which can be used to analyze the customer's focus.

[0148] S44: The field of view is determined by the direction of the gaze and the orientation of the head.

[0149] The field of view refers to the spatial area that a customer can observe with their head facing the same direction as their line of sight.

[0150] By combining the head orientation and line of sight with the preset physiological characteristics of human vision (such as a horizontal field of vision of approximately 120°-160° and a vertical field of vision of approximately 60°-80°), the area that the customer can cover in the current posture is obtained as the field of vision range.

[0151] The preset physiological characteristic parameters are pre-set by technicians based on conclusions of human visual physiology.

[0152] S45: Obtain the target reference point based on the field of view and the reference point it passes through.

[0153] The target reference point refers to the store that matches the surrounding stores when the customer turns their head and receives the customer's active attention.

[0154] By analyzing the overlap between the customer's field of view and a reference point, it is determined whether there is any intersection between the field of view and the store's floor area. If the customer's field of view covers an area of ​​a store, then that store is used as the target reference point.

[0155] S46: Obtain consumer characteristics through the target reference point.

[0156] Consumer characteristics refer to the typical characteristics of the corresponding target consumer group, including age group, gender orientation, spending power, and purchasing preferences.

[0157] By using target reference points and combining the store consumer characteristic information stored in the reference information, the consumer characteristics corresponding to each target reference point are retrieved to form a set of consumer characteristics.

[0158] S47: Determine the detection category by combining consumer characteristics with the detection model.

[0159] The system matches the characteristics of the consumer subject with the individual customer characteristics extracted from the detection model established in S29 to determine whether the customer characteristics are consistent with the target customer group characteristics of the store being monitored, and uses the consistent characteristics as the detection category.

[0160] Methods for updating labeled features include: S48: Retrieve historical reference types from user information.

[0161] Historical reference types refer to store content (including store name, store type, target characteristics, etc.) that customers have made purchases in the past.

[0162] By retrieving customers' historical consumption records from the user information collected by S10, the store content corresponding to each consumption (including store name, store type, target characteristics, etc.) is extracted to form a historical reference type, which is used for subsequent analysis of customers' consumption habits and periodic patterns.

[0163] S49: Obtain the actual output cycle and the reference output cycle based on the historical reference type.

[0164] Actual output cycle refers to the actual time interval between customer inspections of stores based on historical reference types.

[0165] The baseline output cycle refers to the average detection time interval for a specific type of store, calculated based on historical customer consumption data.

[0166] By using historical reference types, the time interval between two consecutive purchases is calculated to obtain the actual output cycle. Simultaneously, the average purchase cycle of all individuals for the same type of product is retrieved from the cycle database as the baseline output cycle.

[0167] The periodic database stores the average purchase cycle of all consumers for various types of goods. The parameters in the periodic database are preset by those skilled in the art based on actual conditions.

[0168] S50: The marked output cycle is obtained based on the consistency between each actual output cycle and the reference output cycle.

[0169] The marked output cycle refers to the actual output cycle after removing abnormal consumption cycles.

[0170] By extracting the two periods with the largest difference from the actual output period, calculating the absolute value of the difference between these two periods and the reference output period, and taking the actual output period corresponding to the period with the smaller absolute value as the marked output period.

[0171] S51: Obtain the abnormal output period based on the marked output period and the actual output period.

[0172] Abnormal output cycle refers to the actual output cycle that deviates significantly from the marked output cycle and may reflect abnormal consumption behavior or special consumption scenarios (such as the existence of purchasing agents).

[0173] The deviation rate is calculated by comparing the actual output cycle with the marked output cycle. If the deviation of the actual output cycle from the marked output cycle exceeds a preset threshold (e.g., ±30%), the actual output cycle is marked as an abnormal output cycle.

[0174] S52: Update the labeling features based on the abnormal output cycle and historical reference type.

[0175] Under the premise of abnormal output cycles, the features of products in the historical reference type are obtained, thereby reducing the impact of the product on the consumer's interested products and achieving the purpose of updating the labeled features.

[0176] Methods for updating labeled features also include: S53: Collect other detection data through user information.

[0177] Other detection data refers to data collected from publicly available platforms (such as social media, e-commerce platforms, membership systems, review platforms, etc.) after obtaining the account information of customers based on their user information, and after receiving a pop-up reminder and consent.

[0178] After obtaining consent from the system, the system logs into the corresponding platform to collect publicly available customer data as additional detection data. This data is used to supplement the information dimensions of offline consumer behavior analysis and improve the completeness and accuracy of the final output information.

[0179] S54: Obtain other text content and other operation data based on other detection data.

[0180] Other text content refers to text, images, videos, and other content information that customers publicly post on the platform.

[0181] Other operational data refers to records of customer interactions on the platform, such as likes, comments, reposts, browsing, favorites, and searches.

[0182] A following list refers to the collection of accounts, brands, stores, and other objects that a customer follows on a platform.

[0183] By retrieving three types of information from other detection data—other text content, other operational data, and the follow list—we can then analyze customers' interests, preferences, and consumption tendencies.

[0184] S55: Obtain other reference ranges based on other text content and other operational data.

[0185] Other reference ranges refer to the range of product categories, brands, features, etc. that customers are interested in, inferred from their content postings and behavioral interactions on the platform.

[0186] Text analysis and image recognition are performed on other text content to extract product keywords, brand tags, and category information mentioned or displayed. Statistical analysis is conducted on other operational data to identify product types with high-frequency customer interaction (picking up and viewing the same type of product more than 3 times) and long-term browsing (browsing the product for more than 10 seconds). The extracted information is then integrated to form other reference ranges.

[0187] S56: Obtain the target feature range based on reference information and target features.

[0188] The target feature range refers to the product categories corresponding to the goods sold by each store stored in the reference information.

[0189] By retrieving the product category distribution information of stores from the reference information, the product categories covered by the reference points in the mall (such as men's clothing, women's clothing, children's clothing, footwear, bags, cosmetics, digital products, catering, etc.) are extracted to form the target feature range.

[0190] S57: By comparing the target feature range with other reference ranges, if the target feature range is within other reference ranges, the marking range is obtained by comparing the target feature range with other reference ranges.

[0191] The labeled range refers to the set of ranges formed by extracting the locations of matching product categories and their corresponding stores when the target feature range of a store is covered by other reference ranges of customers.

[0192] By comparing other reference ranges with the target feature range: if a certain category in the target feature range is completely within other reference ranges (i.e., the category of the product purchased by the customer belongs to the category of the product they are interested in), then the category and its corresponding store information are extracted to form a marking range, which is used to accurately mark the stores and products that the customer may be interested in later.

[0193] S58: Obtain the target features based on the labeling range and update the labeling features.

[0194] Target features refer to product-related features (such as product category, color, shape, price range, brand, etc.) that customers show clear interest in, extracted from the target range.

[0195] By defining the labeling range and combining it with the target feature information from the reference information, specific features matching the product category are extracted to form the labeling target features. These labeling target features are then merged and updated with the existing labeling features from steps S21 and S52.

[0196] S59: If the target feature range is not within other reference ranges, update the labeled features based on other reference ranges.

[0197] If the target feature range is not within other reference ranges, it indicates that the store's product categories fail to cover customers' online interests and preferences. In this case, the category features of the products that the customer is interested in online (such as cases where the customer has made payments) are extracted as supplementary labeling features and recorded.

[0198] Methods for updating labeled features also include: S60: Infer target features based on other detection data.

[0199] Inferred target features refer to the features of products that customers may be interested in or potentially purchase, inferred from publicly available data on customer information collection platforms.

[0200] By combining other detection data, other text content, other operational data, and the watchlist, and through methods such as text analysis, image recognition, and behavioral feature mining, we can extract features such as the category, color, shape, brand, price range, and style of the products that customers are interested in, and form inferred target features.

[0201] S61: Determine whether the target features are consistent by comparing the inferred target features, historical reference types, and target features.

[0202] The matching result is determined by cross-comparing the inferred target features, historical reference types, and target features: if the inferred target features match the historical reference types and target features, and all three have common characteristics, the matching result is determined to be consistent; if any one or all three are inconsistent, the matching result is determined to be inconsistent.

[0203] S62: If they match, retrieve the total output information for the abnormal output period from the user information.

[0204] Total output information refers to the total amount of money spent by customers in the corresponding store during the abnormal output period.

[0205] When the matching result is determined to be consistent, all consumption records of the corresponding consumption store within the abnormal output period are retrieved from the user information collected by S10 according to the abnormal output period, and the output information is summarized to obtain the total output information.

[0206] S63: Update the label features based on the total output information.

[0207] The total consumption amount and the target feature corresponding to the single consumption amount are obtained by counting the single consumption amount within the abnormal consumption cycle obtained in S51, and then removed from the data corresponding to the labeled feature to obtain a new labeled feature.

[0208] S64: If there is a discrepancy, then the abnormal target features are obtained based on the inferred target features, historical reference types, and target features.

[0209] Abnormal target features refer to new product features (shape, color, etc.) obtained when the online inferred target features do not match the actual offline consumption behavior.

[0210] When the matching result in S61 is determined to be inconsistent, the features of the other two inconsistent products (inferred target features, historical reference type and target features) will be regarded as abnormal target features.

[0211] S65: Update the labeling features based on the abnormal target characteristics and the abnormal output cycle.

[0212] By identifying the characteristics of abnormal consumer goods, we can remove the product data corresponding to the identified characteristics and obtain new identified characteristics, thereby reducing the impact of abnormal behavior on the accuracy of data collection and integration.

[0213] Based on the same inventive concept, embodiments of the present invention provide a data acquisition system based on dwell depth, comprising: The acquisition module is used to obtain user location and user information; A memory used to store a program for a data acquisition method based on dwell depth; The processor is used to load and execute programs stored in memory.

[0214] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0215] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A data acquisition method based on dwell depth, characterized in that, include: Collect user location and user information; The dwell time is determined based on the user's location and a preset reference point; Retrieve detection time, target features, and actual output information from user information; The actual location is marked based on the detection time and the user's location; The number of visits was obtained based on the actual location and reference points; The final output information is obtained by combining dwell time, number of visits, reference points, actual output information and target characteristics, and a report is uploaded. The methods for correcting the final output information include: The remaining location is obtained based on the user's location and the actual location; The remaining path and remaining dwell time are obtained by comparing the remaining location with the preset reference range; The number of marked references is obtained based on the remaining path and the reference range; The detection level is determined by comparing the remaining dwell time with the preset baseline dwell time range. The range of marked categories is obtained by combining the remaining locations, detection levels, and preset reference information; The tag features are obtained based on the tag category range and reference information; The final output information is obtained by combining the detection level, the number of reference markers, and the features of the reference points and markers, and then uploaded.

2. The data acquisition method based on dwell depth according to claim 1, characterized in that, The methods for correcting the remaining position include: Retrieve baseline path features from reference information; The selected path is determined based on the reference range and the remaining path. Based on the baseline path features and the selected path, the overlapping path and the final path are derived. The detection and placement image is obtained by collecting image detection information based on the final path. The actual baseline path is obtained by detecting the placement of images and the final path; The detection positions are obtained by comparing the actual baseline path with the overlapping path, and the detection positions are removed from the remaining positions to update the remaining positions.

3. The data acquisition method based on dwell depth according to claim 2, characterized in that, The methods for correcting the final output information also include: Collect detection images of the user's location and build a detection model based on the detection images and preset human features; The detection interval is obtained based on the user's location; The aggregation range is obtained by comparing the detection interval with the preset row spacing; The aggregation range is updated at a preset detection time, and the detection range is obtained by comparing the aggregation range before and after the update. The detection category is obtained by using the detection model and the detection range; The final output information is updated based on the detection category and reference point.

4. The data acquisition method based on dwell depth according to claim 3, characterized in that, Methods for updating the final output information include: The reference points are obtained based on the actual location and the remaining path. The reference category is obtained based on the reference information and the reference points visited. The target category is obtained by comparing the detected category with the benchmark reference type; The walking speed is obtained by using the remaining position and the reference range; Retrieve the reference type from the reference category; The marker's movement speed is obtained based on the detection category; The detection reference point is obtained by comparing the walking speed with the marker speed and combining the reference range; The final output information is updated based on the target category and the detection reference point.

5. The data acquisition method based on dwell depth according to claim 4, characterized in that, The methods for verifying the detection category include: Retrieve the gaze direction and head orientation from the detection model; The field of view is determined by the direction of the gaze and the orientation of the head. The target reference point is obtained based on the field of view and the reference points it passes through; The characteristics of the consumer subject are obtained through the target reference point; The detection category is determined by combining the characteristics of the consumer group with the detection model.

6. The data acquisition method based on dwell depth according to claim 5, characterized in that, Methods for updating labeled features include: Retrieve historical reference types from user information; The actual output cycle and the baseline output cycle are obtained based on the historical reference type; The marked output cycle is obtained based on the consistency between each actual output cycle and the reference output cycle; The abnormal output period is obtained based on the marked output period and the actual output period; Update the labeling features based on the abnormal output cycle and historical reference type.

7. The data acquisition method based on dwell depth according to claim 6, characterized in that, Methods for updating labeled features also include: Collect other detection data using user information; Other text content and other operation data are obtained based on other detection data; Other reference ranges are obtained based on other text content and other operational data; The target feature range is obtained based on reference information and target features; By comparing the target feature range with other reference ranges, if the target feature range is within other reference ranges, the marking range is obtained by comparing the target feature range with other reference ranges. Obtain the target features based on the labeling range and update the labeling features; If the target feature range is not within other reference ranges, update the labeled features based on other reference ranges.

8. The data acquisition method based on dwell depth according to claim 7, characterized in that, Methods for updating labeled features also include: Target features are inferred based on other detection data; The consistency between the inferred target characteristics, historical reference types, and target characteristics is determined by comparing them. If they match, retrieve the total output information for the abnormal output period from the user information. Update the labeled features based on the total output information; If there is a discrepancy, the abnormal target features are obtained based on the inferred target features, historical reference types, and target features. The labeling features are updated based on the characteristics of the abnormal target and the abnormal output cycle.

9. A data acquisition system based on dwell depth, characterized in that, include: The acquisition module is used to obtain user location and user information; A memory for storing a program that implements a data acquisition method based on dwell depth as described in any one of claims 1 to 8; The processor is used to load and execute programs stored in memory.

Images