Commodity recommendation method based on customer attribute analysis

Abstract

The invention relates to the technical field of computers, and discloses a customer attribute analysis-based commodity recommendation method, which comprises the following steps of: performing frame segmentation on a video stream when a customer enters a store to obtain a single-frame image, and preprocessing the single-frame image; obtaining corresponding customer attributes according to the single-frame image, wherein the customer attributes comprise facial features and clothing features; acquiring behavior data of the corresponding customer based on a moving target detection technology, and acquiring interest points of the customer according to the behavior data; and constructing a customer portrait according to the customer attributes and the behavior data, generating a preliminary recommendation list through a preset recommendation strategy, and respectively endowing the behavior data and the customer attributes with different weights so as to recommend commodities matched with the customer portrait. According to the method, the facial features and the clothing styles of the customers are analyzed, and the behavior data of the customers in the shop are combined to construct the customer portraits, so that accurate recommendation of the commodities is realized.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a commodity recommendation method based on customer attribute analysis. Background Art

[0002] With the rapid development of the retail industry and the advancement of technology, how to use modern information technology to improve customers' shopping experience and merchants' service quality has become an important issue. Traditional product recommendation methods often rely on the experience and personal preferences of sales staff. This method has certain limitations and is difficult to meet customers' personalized needs.

[0003] In recent years, with the development of computer vision technology and big data analysis, product recommendation systems based on customer attribute analysis have gradually become a research hotspot. However, most recommendation systems in the existing technology only make recommendations based on customers' historical purchase records or browsing behaviors, while ignoring customers' real-time preferences and on-site behavior data, resulting in inaccurate product recommendations.

[0004] Therefore, a product recommendation method based on customer attribute analysis is needed to solve the problems in existing recommendation systems and improve customers' shopping experience. Summary of the invention

[0005] In view of this, the present invention proposes a product recommendation method based on customer attribute analysis, which constructs a customer portrait by analyzing the customer's facial features and clothing style and combining the customer's in-store behavior data, aiming to solve the problem of inaccurate product recommendations in current technology.

[0006] The present invention proposes a product recommendation method based on customer attribute analysis, comprising:

[0007] Perform frame segmentation on the video stream when customers enter the store, obtain single-frame images and perform preprocessing;

[0008] Acquire corresponding customer attributes according to the single-frame image, wherein the customer attributes include facial features and clothing features;

[0009] Acquire the behavior data of the corresponding customer based on the motion target detection technology, and acquire the customer's interest points according to the behavior data;

[0010] A customer profile is constructed based on the customer attributes and the behavior data, a preliminary recommendation list is generated through a preset recommendation strategy, and different weights are assigned to the behavior data and the customer attributes to recommend products that match the customer profile.

[0011] Preferably, the corresponding customer attributes are obtained according to the single-frame image, and the customer attributes include facial features and clothing features, specifically:

[0012] Detect the customer's face based on image processing technology, and use deep learning models to extract and classify the detected facial areas to determine the customer's age range and gender;

[0013] The color distribution and style characteristics of the customer's clothing are extracted, and the customer's clothing is classified.

[0014] Preferably, the detecting of the customer's face based on image processing technology further comprises:

[0015] After the facial area is detected, the sensitive area is blurred using privacy protection technology to retain sufficient facial features for subsequent attribute analysis.

[0016] Preferably, the extracting of the color distribution and style characteristics of the customer's clothing and classifying the customer's clothing is specifically as follows:

[0017] Determine the color attributes of the corresponding customer according to the clothing color distribution, where the color attributes include cool color outfits, warm color outfits, neutral color outfits, and other outfits;

[0018] The clothing style of the corresponding customer is determined according to the style characteristics, and the clothing style includes business style, casual style, sports style and other styles.

[0019] Preferably, the color attribute of the corresponding customer is determined according to the clothing color distribution, specifically:

[0020] When the cool colors on the customer's clothes exceed the preset threshold, it can be determined that the customer is wearing cool colors, including blue, green and purple;

[0021] When the warm colors on the customer's clothes exceed the preset threshold, it can be determined that the customer is wearing warm colors, which include red, orange and yellow;

[0022] When the neutral colors on the customer's clothes exceed the preset threshold, it can be determined as neutral color matching, and the neutral colors include black, white, gray, brown and beige;

[0023] When the cool colors, the warm colors, and the neutral colors on the customer's clothes do not exceed the preset threshold, it can be determined as other outfits.

[0024] Preferably, the clothing style of the corresponding customer is determined according to the style characteristics, specifically:

[0025] Obtaining style features of the customer based on the single-frame image, and determining the customer's clothing style according to a comparison result of the style features with a preset threshold, wherein the style features include texture contrast, color type, pattern identification, and homogeneity;

[0026] When the texture contrast is low, the color types are few, the pattern logo is simple, and the homogeneity is high, it can be determined as a business style;

[0027] When the texture contrast is high, the color variety is large, the pattern logo is simple, and the homogeneity is low, it can be determined as a casual style;

[0028] When the texture has high contrast, many colors, complex pattern logos, and high homogeneity, it can be determined as a sports style;

[0029] When the clothing style does not belong to business style, casual style and sports style, it can be determined as other styles.

[0030] Preferably, the method of acquiring the corresponding customer's behavior data based on the moving target detection technology and acquiring the customer's interest points according to the behavior data is specifically as follows:

[0031] Track the movement path of customers in the store through motion target detection technology, and analyze the customers' points of interest based on their stay time and location information;

[0032] If the stay time is more than 10 seconds but less than 20 seconds, the point of interest is determined to be a primary point of interest;

[0033] If the stay time is more than 20 seconds but less than 30 seconds, the point of interest is determined to be an intermediate point of interest;

[0034] If the stay time exceeds 30 seconds, the point of interest is determined to be a high-level point of interest.

[0035] Preferably, the preliminary recommendation list is generated by a preset recommendation strategy, and the recommendation strategy assigns different weights to the behavior data and the customer attributes, specifically:

[0036] The recommendation strategy adopts a linear weighted method to obtain the types of goods required by the customer based on the behavior data, and determines the types of goods that meet the customer profile in combination with the customer attributes to obtain a preliminary recommendation list;

[0037] The weight of the behavior data in the recommendation list is greater than the weight of the customer attributes.

[0038] Preferably, it also includes:

[0039] assigning corresponding weights to the facial features and the clothing features based on the recommendation strategy, wherein the weight of the facial features is greater than the weight of the clothing features;

[0040] assigning corresponding weights to the age range and the gender based on the recommendation strategy, wherein the weight of the gender in the facial features is greater than the weight of the age range;

[0041] Based on the recommendation strategy, corresponding weights are assigned to the color attributes and the clothing style, and the weight of the clothing style in the clothing features is greater than the weight of the color attributes.

[0042] Preferably, it also includes:

[0043] Based on the recommendation strategy, the corresponding weight of each of the interest points is assigned, the weight of the high-level interest points in the behavior data is greater than the weight of the intermediate interest points, and the weight of the intermediate interest points in the behavior data is greater than the weight of the low-level interest points.

[0044] The embodiment of the present application discloses a method for recommending products based on customer attribute analysis. This method can more accurately understand the customer's preferences by analyzing the customer's facial features and clothing style, combined with the customer's in-store behavior data, and thus provide product recommendations that better meet the customer's needs; in the process of obtaining the customer's facial features, the sensitive area is blurred through privacy protection technology, which not only ensures the customer's privacy security, but also retains enough facial features for subsequent analysis; by giving different weights to different types of customer attributes and behavior data, the recommendation strategy is made more flexible and targeted, and the accuracy of the recommendation results is improved; through accurate product recommendations, the time for customers to find the desired products is shortened, and the customer's shopping experience and satisfaction are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0046] Figure 1 A flowchart of a method for recommending products based on customer attribute analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0048] like Figure 1As shown, in some embodiments of the present application, this embodiment provides a product recommendation method based on customer attribute analysis. Specifically, the method includes the following steps:

[0049] Step S101: Segment the video stream of a customer entering a store to obtain a single frame image and perform preprocessing.

[0050] As mentioned above, this step aims to extract useful static image frames from a continuous video stream as the basis for subsequent analysis. The video stream captured by the camera in real time contains a series of continuous image frames. By performing frame segmentation on the video stream, the video can be decomposed into a series of separate image frames. After that, each frame of the image is preprocessed, including but not limited to operations such as resizing, color space conversion, and noise removal, to improve the efficiency and accuracy of subsequent feature extraction.

[0051] For example, suppose the camera captures 30 frames per second, and the system starts recording a video stream when a customer enters the store. For analysis purposes, the system chooses to extract 1 frame per second for subsequent processing. For example, the first frame is extracted at the first second, the second frame at the second second, and so on. For each frame, the system first converts it from the RGB color space to a grayscale image to reduce the amount of computation and increase processing speed. Next, the system resizes the image to 512x512 pixels, which ensures that all images have the same size for subsequent feature extraction and recognition.

[0052] It should be noted that, based on the above-mentioned scheme, those skilled in the art may also adopt an image enhancement scheme, that is, when the light in the store is dim or uneven, image enhancement technology (such as histogram equalization or local contrast enhancement) may be used to improve the image quality, so that the facial features and clothing details of the customers are more clearly visible; a background subtraction scheme may be adopted, that is, background subtraction technology (such as frame difference or background modeling method) may be used to remove the unchanging background part from the image, and only the moving target (i.e., the customer) may be retained, thereby reducing unnecessary computational load and improving the accuracy of feature extraction. The above optional schemes all fall within the scope of protection of this application.

[0053] Step S102: Acquire corresponding customer attributes according to the single-frame image, wherein the customer attributes include facial features and clothing features.

[0054] As mentioned above, the system obtains the customer's facial features and clothing features by analyzing a single frame of image. Facial features mainly include the customer's age range, gender, etc.; clothing features cover the customer's clothing color distribution and style characteristics. By extracting these features, basic data can be provided for subsequent customer portrait construction and product recommendations.

[0055] For example, assume that the system has extracted a frame of image from a video stream. Next, the system first uses facial detection technology (such as Haar feature cascade classifier or deep learning-based detection model) to locate the customer's facial area. Once the face is detected, the system will use a pre-trained age and gender classification model (such as a convolutional neural network-based model) to estimate the customer's age range and gender. At the same time, the system will also analyze the customer's clothing, extract the color distribution characteristics and texture features of the clothing, use tools such as color histograms and texture descriptors, and combine deep learning models to classify clothing styles.

[0056] It should be noted that technical personnel in this field may also adopt an advanced feature extraction scheme based on the above scheme, that is, in addition to the basic age and gender classification, more complex deep learning models may be used to extract more detailed facial features, such as expressions, skin color, etc. These features help to further refine the customer portrait; adopt a multi-view facial detection scheme, that is, considering that the facial features of the customer may be different at different angles, multi-view facial detection technology may be used, which can accurately detect and analyze facial features even when the customer is at a non-frontal angle; adopt a multimodal feature fusion scheme, that is, in addition to visual features, other modal data (such as the customer's voice features, gait features, etc.) can also be combined, and through multimodal feature fusion technology, the dimensions of customer attributes can be further enriched and the accuracy of recommendations can be improved. The above optional schemes all fall within the scope of protection of this application.

[0057] Step S103: obtaining the behavior data of the corresponding customer based on the motion target detection technology, and obtaining the customer's interest points according to the behavior data.

[0058] As mentioned above, the system uses motion target detection technology to track and record customers' movement paths, dwell time, and other related behavior data in specific scenarios. Based on these behavior data, the system further analyzes customers' shopping preferences and interests, such as which products customers spend more time in front of, or the frequency of customers' activities in specific areas of the store.

[0059] For example, suppose that in a retail store environment, the system captures a series of customer behaviors through a camera mounted on the ceiling. First, the system uses motion target detection techniques such as background subtraction or optical flow to identify the customer's movement trajectory. For example, when a customer enters the camera's field of view, the system begins to track the customer's movement path. If a customer stays in an area displaying specific products for a long time, the system will record this behavior and believe that the customer may be interested in this type of product. In addition, the system can also record the customer's walking speed, direction changes and other behavioral characteristics, which are crucial for understanding the customer's behavior patterns.

[0060] It should be noted that, based on the above scheme, those skilled in the art may also adopt an abnormal behavior detection scheme, that is, use machine learning algorithms (such as cluster analysis or outlier detection algorithms) to identify abnormal behaviors of customers, such as sudden acceleration or deceleration, wandering, etc. These abnormal behaviors may indicate that customers are particularly interested in or confused about certain products; adopt a customer emotion analysis scheme, that is, combine facial expression recognition technology to analyze the emotional reactions of customers in front of different products, such as happiness, disappointment, etc., which helps to more deeply understand the customers' interests and purchasing intentions; adopt a customer behavior prediction scheme, that is, based on historical behavior data, use time series analysis or prediction models (such as LSTM networks) to predict the customer's future movement trajectory and interests, so as to prepare for product recommendations or promotional activities in advance. The above optional schemes all fall within the scope of protection of this application.

[0061] Step S104: construct a customer profile based on the customer attributes and the behavior data, generate a preliminary recommendation list through a preset recommendation strategy, assign different weights to the behavior data and the customer attributes, so as to recommend products that match the customer profile.

[0062] As mentioned above, the system builds a comprehensive customer profile based on previously collected customer attributes (such as age, gender, etc.) and behavioral data (such as customer shopping habits, points of interest, etc.). Then, the system generates a preliminary product recommendation list based on the customer profile through a preset recommendation algorithm or strategy. During this process, the system assigns different weights to different behavioral data and customer attributes to ensure that the recommended products are highly matched with the customer's actual needs and preferences.

[0063] For example, suppose the system has collected customer behavior data through devices such as cameras, and obtained some of the customer's personal information through other channels (such as member registration information). For example, a young female customer often stops at the cosmetics counter, and she is also a loyal member. The system can build her customer portrait based on information such as her gender, age, and the time she stays at the cosmetics counter. Then, the system can set a recommendation strategy, such as giving priority to hot-selling cosmetics, and giving higher weights based on the customer's stay time at the cosmetics counter, while considering factors such as her age and gender to give appropriate weights. Finally, the system generates a recommendation list containing hot-selling cosmetics, and considering that this young female customer may be more interested in skin care products, more skin care products will be included in the recommendation list.

[0064] It should be noted that, based on the above-mentioned scheme, those skilled in the art may also adopt a dynamic weight adjustment scheme, that is, dynamically adjust the weights of behavioral data and customer attributes according to real-time customer feedback (such as click-through rate, purchasing behavior, etc.) to continuously optimize the recommendation effect; adopt a context-aware scheme, that is, adjust the recommendation strategy in combination with the customer's current context (such as season, weather, time, etc.), such as increasing the recommendation weight of warm clothing in winter; adopt a collaborative filtering scheme, that is, use a collaborative filtering algorithm (such as user-based collaborative filtering or item-based collaborative filtering) to recommend products that are liked by other customers who are similar to the customer portrait. The above optional schemes all fall within the scope of protection of this application.

[0065] In some embodiments of the present application, in order to obtain basic attribute information of customers and provide basic data support for subsequent tasks such as customer portrait construction and personalized recommendation, the corresponding customer attributes are obtained according to the single-frame image, and the customer attributes include facial features and clothing features, specifically:

[0066] Detect the customer's face based on image processing technology, and use deep learning models to extract and classify the detected facial areas to determine the customer's age range and gender;

[0067] The color distribution and style characteristics of the customer's clothing are extracted, and the customer's clothing is classified.

[0068] As described above, the facial region in a single frame image is identified using image processing techniques. The face is located by preprocessing the image (including grayscale, noise removal, etc.) and then applying a facial detection algorithm (such as a Haar feature cascade classifier or a deep learning-based method). A deep learning model (such as a convolutional neural network (CNN)) is applied to the detected facial region for feature extraction, aiming to extract key features from the facial image that are useful for age and gender recognition. The extracted facial features are classified using the trained model to predict the approximate age range and gender of the customer. The color of the clothes worn by the customer is analyzed by calculating the color histogram at the pixel level to determine its main color distribution. The style and pattern of the clothes are analyzed to extract style features. For example, a convolutional neural network or other computer vision technology can be used to identify specific patterns, textures, and other features on the clothes. Combining color distribution and style features, a pre-trained classification model is used to classify the customer's clothing type, such as casual wear, business wear, etc.

[0069] In some embodiments of the present application, in order to comply with the requirements of data protection and privacy regulations and ensure that the system can accurately extract useful information from the image, thereby achieving accurate analysis of customer attributes, the customer's face is detected based on image processing technology, and further includes:

[0070] After the facial area is detected, the sensitive area is blurred using privacy protection technology to retain sufficient facial features for subsequent attribute analysis.

[0071] As mentioned above, first, a facial detection algorithm, such as a Haar feature-based classifier or a more advanced deep learning method, such as a convolutional neural network (CNN), is used to accurately frame the facial contours and identify and locate the facial area in the image. In order to comply with privacy protection regulations and respect personal privacy, sensitive information needs to be processed after the facial area is located. Specifically, sensitive areas of the face (such as the eyes or mouth, etc.) are blurred by applying Gaussian blur, pixelation, or other image processing techniques to ensure that sensitive information cannot be easily identified. By fine-tuning the degree of blur, not only privacy is ensured, but also key facial features can be correctly identified and extracted. Finally, the processed facial images are fed into a deep learning model for feature extraction and classification. These deep learning models, which are trained to recognize facial features and estimate attributes such as age range and gender, can maintain a high accuracy rate even when parts of the facial area are blurred.

[0072] In some embodiments of the present application, in order to accurately extract the color distribution and style characteristics of the customer's clothing, and classify the customer's clothing into corresponding color attributes and style categories based on these characteristics, to provide support for subsequent personalized product recommendations. The color distribution and style characteristics of the customer's clothing are extracted, and the customer's clothing is classified, specifically:

[0073] Determine the color attributes of the corresponding customer according to the clothing color distribution, where the color attributes include cool color outfits, warm color outfits, neutral color outfits, and other outfits;

[0074] The clothing style of the corresponding customer is determined according to the style characteristics, and the clothing style includes business style, casual style, sports style and other styles.

[0075] As described above, first, the color information is extracted from the customer's clothing image by calculating the color histogram to represent the distribution of different colors in the clothing. Then, the main color composition of the customer's clothing is analyzed based on the color histogram. According to the color characteristics of the main colors, the customer's clothing is divided into color attributes such as cool color matching, warm color matching, neutral color matching, and other matching. Style features are extracted from the customer's clothing image, and these features include patterns, textures, styles, etc. Based on the extracted style features, the customer's clothing style is classified using a pre-trained model (such as a classification model based on deep learning), and the customer's clothing style is divided into business style, casual style, sports style, and other styles.

[0076] In some embodiments of the present application, in order to accurately determine the color attribute of the customer's clothes according to the color distribution of the clothes, so as to provide support for subsequent personalized product recommendations. According to the color distribution of the clothes, the color attribute of the corresponding customer is determined, specifically:

[0077] When the cool colors on the customer's clothes exceed the preset threshold, it can be determined that the customer is wearing cool colors, including blue, green and purple;

[0078] When the warm colors on the customer's clothes exceed the preset threshold, it can be determined that the customer is wearing warm colors, which include red, orange and yellow;

[0079] When the neutral colors on the customer's clothes exceed the preset threshold, it can be determined as neutral color matching, and the neutral colors include black, white, gray, brown and beige;

[0080] When the cool colors, the warm colors, and the neutral colors on the customer's clothes do not exceed the preset threshold, it can be determined as other outfits.

[0081] As described above, color information is extracted from the customer's clothing images and a color histogram is calculated to understand the distribution of different colors in the clothing. The proportion of cool colors (blue, green, purple, etc.) in the color histogram is counted, and the preset threshold is set to 60%, that is, when the proportion of cool colors exceeds 60%, the customer's clothing is judged to be cool-colored; the proportion of warm colors (red, orange, yellow, etc.) in the color histogram is counted, and the preset threshold is set to 60%, that is, when the proportion of warm colors exceeds 60%, the customer's clothing is judged to be warm-colored; the proportion of neutral colors (black, white, gray, brown, beige, etc.) in the color histogram is counted, and the preset threshold is set to 60%, that is, when the proportion of neutral colors exceeds 60%, the customer's clothing is judged to be neutral-colored; when the proportion of cool colors, warm colors and neutral colors on the customer's clothes are all lower than the preset threshold of 60%, it means that the color distribution of the customer's clothes is more complex and is not easy to be classified into one of the above three color systems. At this time, the customer's clothes are judged to be other outfits.

[0082] In some embodiments of the present application, in order to accurately classify a customer's clothing into business style, casual style, sports style or other styles according to the style characteristics of the customer's clothing, so as to provide support for subsequent personalized product recommendations. According to the style characteristics, the clothing style of the corresponding customer is determined, specifically:

[0083] Obtaining style features of the customer based on the single-frame image, and determining the customer's clothing style according to a comparison result of the style features with a preset threshold, wherein the style features include texture contrast, color type, pattern identification, and homogeneity;

[0084] When the texture contrast is low, the color types are few, the pattern logo is simple, and the homogeneity is high, it can be determined as a business style;

[0085] When the texture contrast is high, the color variety is large, the pattern logo is simple, and the homogeneity is low, it can be determined as a casual style;

[0086] When the texture has high contrast, many colors, complex pattern logos, and high homogeneity, it can be determined as a sports style;

[0087] When the clothing style does not belong to business style, casual style and sports style, it can be determined as other styles.

[0088] As mentioned above, relevant features of the customer's clothing are extracted from a single frame image, including but not limited to texture contrast, color type, pattern identification, and homogeneity. Texture contrast includes analyzing the texture features of clothing, such as the complexity of the texture, the change of the texture, etc.; color type includes counting the number of colors that appear in clothing and understanding color diversity; pattern identification includes identifying patterns or logos on clothing, such as brand logos, special graphics, etc.; homogeneity includes evaluating the consistency of clothing in color, texture, etc. The extracted style features are compared with the preset threshold to determine the customer's clothing style. When the texture contrast is low, the color types are few, the pattern logos are simple and the homogeneity is high, it means that the clothes are usually simple, single in color, simple in pattern, and have high consistency overall, which can be determined as a business style; when the texture contrast is high, the color types are many, the pattern logos are simple and the homogeneity is low, it means that the clothes are rich in color, complex in texture, simple in pattern, and the color, texture and other features are unevenly distributed on the clothes, which can be determined as a casual style; when the texture contrast is high, the color types are many, the pattern logos are complex and the homogeneity is high, it means that the clothes have obvious sports elements, such as complex pattern logos, bright colors, consistent textures, etc., which can be determined as a sports style; when the clothing style does not belong to business style, casual style and sports style, that is, when the style characteristics do not meet the conditions of any of the above styles, it can be determined as other styles.

[0089] In some embodiments of the present application, in order to accurately analyze the customer's points of interest based on the customer's in-store behavior data (such as movement path and dwell time), and divide them into different levels of points of interest according to different dwell times, so as to provide support for subsequent personalized product recommendations. Based on the motion target detection technology, the corresponding customer's behavior data is obtained, and the customer's points of interest are obtained according to the behavior data, specifically:

[0090] Track the movement path of customers in the store through motion target detection technology, and analyze the customers' points of interest based on their stay time and location information;

[0091] If the stay time is more than 10 seconds but less than 20 seconds, the point of interest is determined to be a primary point of interest;

[0092] If the stay time is more than 20 seconds but less than 30 seconds, the point of interest is determined to be an intermediate point of interest;

[0093] If the stay time exceeds 30 seconds, the point of interest is determined to be a high-level point of interest.

[0094] As mentioned above, firstly, the movement of customers in the store is identified by using motion target detection technology, such as background subtraction or optical flow method; then the movement trajectory of customers is captured by continuous video frames, and the time customers stay at various locations is recorded; combined with the layout of the store and the layout of the camera, the specific location information of customers in the store is obtained. The customer's stay time in the store is analyzed, and the length of time customers stay in different areas is recorded; the customer's points of interest are determined according to the length of time customers stay and the location; different thresholds of stay time are set to distinguish the degree of customer interest. When a customer stays in a certain area for more than 10 seconds but less than 20 seconds, it indicates that the customer has a certain interest in the goods in the area, but may need more information to make a purchase decision, and the area can be judged as the customer's primary point of interest; when a customer stays in a certain area for more than 20 seconds but less than 30 seconds, it indicates that the customer has a high interest in the goods in the area and may be carefully considering whether to buy, and the area can be judged as the customer's intermediate point of interest; when a customer stays in a certain area for more than 30 seconds, it indicates that the customer is very interested in the goods in the area and may be ready to buy or have made a purchase decision, and the area can be judged as the customer's advanced point of interest.

[0095] In some embodiments of the present application, in order to more accurately reflect the needs and interests of customers and improve the effectiveness and satisfaction of recommendations, a preliminary recommendation list is generated through a preset recommendation strategy, and the recommendation strategy assigns different weights to the behavior data and the customer attributes, specifically:

[0096] The recommendation strategy adopts a linear weighted method to obtain the types of goods required by the customer based on the behavior data, and determines the types of goods that meet the customer profile in combination with the customer attributes to obtain a preliminary recommendation list;

[0097] The weight of the behavior data in the recommendation list is greater than the weight of the customer attributes.

[0098] As mentioned above, the linear weighting method is used as the core mechanism of the recommendation algorithm, and a weight coefficient is assigned to each type of data to measure its influence on the final recommendation result. Analyze the customer's behavior data, including but not limited to the customer's movement path and residence time in the store; extract the customer's preference for specific types of goods from the behavior data, such as which areas or products attract more attention from customers; determine the customer's interest level based on the length of time the customer stays in front of different product categories, and infer the type of goods that the customer is most interested in. Collect the customer's basic attribute information, such as age, gender, etc.; use this attribute information to build a customer profile to better understand the customer's consumption habits and personal preferences; combine the customer attributes with the existing customer behavior data to further screen out the types of goods that match the customer profile. Combine the product types obtained from the behavior data with the product types that match the customer attributes; comprehensively consider the product types from these two sources, and sort and screen them according to their respective weights; generate a preliminary recommendation list containing the products that are most likely to attract the customer's interest. Set the weight of the behavior data higher than the weight of the customer attributes, which means that the customer's actual behavior is more important than their basic information. For example, in the linear weighted formula, assuming that the weight of behavioral data is W2 and the weight of customer attributes is W1, then W2>W1. According to this weight distribution principle, behavioral data has a greater impact on the final recommendation results, while customer attributes serve as auxiliary factors to adjust the recommendation results.

[0099] In some embodiments of the present application, in order to make product recommendations more precisely based on the specific characteristics of customers, it also includes:

[0100] assigning corresponding weights to the facial features and the clothing features based on the recommendation strategy, wherein the weight of the facial features is greater than the weight of the clothing features;

[0101] assigning corresponding weights to the age range and the gender based on the recommendation strategy, wherein the weight of the gender in the facial features is greater than the weight of the age range;

[0102] Based on the recommendation strategy, corresponding weights are assigned to the color attributes and the clothing style, and the weight of the clothing style in the clothing features is greater than the weight of the color attributes.

[0103] As mentioned above, in order to more accurately identify the identity of customers and their preferences, the recommendation system needs to collect and analyze the facial features (such as age, gender, etc.) and clothing features (such as clothing style, color matching, etc.) of customers. The weight of facial features is set higher because facial features are usually more stable and can better reflect personal identity, which helps to improve the accuracy of customer identification; although clothing features are also important, they are more variable, so the weight given to them in the recommendation strategy is lower than that of facial features; in practical applications, the recommendation effect can be optimized by adjusting the weight ratio of facial features and clothing features. Gender is given a higher weight in facial features because gender is usually easier to accurately identify from facial features than age range, and gender may be more critical for some product recommendations; although age range is also important reference information, it has a relatively low weight in the recommendation strategy because it may be affected by factors such as changes in facial expressions, resulting in increased difficulty in identification; by adjusting the weights of gender and age range, the recommendation strategy can make more personalized product recommendations for each customer. Clothing style is given a higher weight because it reflects the customer's clothing taste and occasion choice, which is particularly important for recommending suitable products; although color attributes are also an important part of clothing characteristics, their importance is relatively low for the decision-making of the recommendation system; by adjusting the weight ratio between clothing style and color attributes, it can be ensured that the recommended products not only meet the customers' aesthetic preferences, but also meet their actual needs.

[0104] In some embodiments of the present application, in order to improve the accuracy and personalization of recommendations, different weights are assigned to points of interest at different levels.

[0105] Based on the recommendation strategy, the corresponding weight of each of the interest points is assigned, the weight of the high-level interest points in the behavior data is greater than the weight of the intermediate interest points, and the weight of the intermediate interest points in the behavior data is greater than the weight of the low-level interest points.

[0106] As mentioned above, in the recommendation strategy, based on the customer's in-store behavior data, the customer's points of interest are divided into three levels: high-level points of interest, medium-level points of interest, and low-level points of interest. High-level points of interest refer to the areas where customers stay the longest and have the strongest interest, followed by medium-level points of interest, and low-level points of interest are areas where customers have relatively weak interest. In order to more accurately reflect the customer's interest level, the recommendation strategy assigns different weights to these three levels of points of interest. High-level points of interest are given the highest weight because these points of interest represent the strongest interests of customers and are therefore very important for the recommendation strategy; medium-level points of interest are given the second highest weight. Although they are not as strong as high-level points of interest, they can still provide valuable information to help the system understand the customer's interest range; low-level points of interest are given the lowest weight because these points of interest represent areas where customers have weaker interest. Although they still need to be considered, their importance is relatively low. Through this weight distribution, the recommendation system can more accurately recommend products based on customers' points of interest, ensuring that the recommended products are highly matched with customers' interests and preferences.

[0107] Compared with the prior art, the embodiment of the present application discloses a product recommendation method based on customer attribute analysis. By analyzing the customer's facial features and clothing style and combining the customer's in-store behavior data, it is possible to more accurately understand the customer's preferences, thereby providing product recommendations that better meet the customer's needs; in the process of obtaining the customer's facial features, sensitive areas are blurred through privacy protection technology, which not only ensures the customer's privacy security, but also retains sufficient facial features for subsequent analysis; by assigning different weights to different types of customer attributes and behavior data, the recommendation strategy is made more flexible and targeted, and the accuracy of the recommendation results is improved; through accurate product recommendations, the time customers spend looking for the products they need is shortened, thereby improving the customer's shopping experience and satisfaction.

[0108] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0109] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0110] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does 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 product recommendation method based on customer attribute analysis, characterized in that: include: Perform frame segmentation on the video stream when customers enter the store, obtain single-frame images and perform preprocessing; Acquire corresponding customer attributes according to the single-frame image, wherein the customer attributes include facial features and clothing features; Acquire the behavior data of the corresponding customer based on the motion target detection technology, and acquire the customer's interest points according to the behavior data; A customer profile is constructed based on the customer attributes and the behavior data, a preliminary recommendation list is generated through a preset recommendation strategy, and different weights are assigned to the behavior data and the customer attributes to recommend products that match the customer profile.

2. The method according to claim 1, characterized in that The corresponding customer attributes are obtained according to the single-frame image, and the customer attributes include facial features and clothing features, specifically: Detect the customer's face based on image processing technology, and use deep learning models to extract and classify the detected facial areas to determine the customer's age range and gender; The color distribution and style characteristics of the customer's clothing are extracted, and the customer's clothing is classified.

3. The method according to claim 2, characterized in that The detecting of the customer's face based on the image processing technology also includes: After the facial area is detected, the sensitive area is blurred using privacy protection technology to retain sufficient facial features for subsequent attribute analysis.

4. The method according to claim 2, characterized in that The step of extracting the color distribution and style characteristics of the customer's clothing and classifying the customer's clothing is as follows: Determine the color attributes of the corresponding customer according to the clothing color distribution, where the color attributes include cool color outfits, warm color outfits, neutral color outfits, and other outfits; According to the style characteristics, the clothing style of the corresponding customer is determined, and the clothing style includes business style, casual style, sports style and other styles.

5. The method according to claim 4, characterized in that Determining the color attribute of the corresponding customer according to the clothing color distribution is specifically as follows: When the cool colors on the customer's clothes exceed the preset threshold, it can be determined that the customer is wearing cool colors, including blue, green and purple; When the warm colors on the customer's clothes exceed the preset threshold, it can be determined that the customer is wearing warm colors, which include red, orange and yellow; When the neutral colors on the customer's clothes exceed the preset threshold, it can be determined as neutral color matching, and the neutral colors include black, white, gray, brown and beige; When the cool colors, the warm colors, and the neutral colors on the customer's clothes do not exceed the preset threshold, it can be determined as other outfits.

6. The method according to claim 4, characterized in that According to the style characteristics, the clothing style of the corresponding customer is determined, specifically: Obtaining style features of the customer based on the single-frame image, and determining the customer's clothing style according to a comparison result of the style features with a preset threshold, wherein the style features include texture contrast, color type, pattern identification, and homogeneity; When the texture contrast is low, the color types are few, the pattern logo is simple, and the homogeneity is high, it can be determined as a business style; When the texture contrast is high, the color variety is large, the pattern logo is simple, and the homogeneity is low, it can be determined as a casual style; When the texture has high contrast, many colors, complex pattern logos, and high homogeneity, it can be determined as a sports style; When the clothing style does not belong to business style, casual style and sports style, it can be determined as other styles.

7. The method according to claim 1, characterized in that The method of obtaining the corresponding customer's behavior data based on the moving target detection technology and obtaining the customer's interest points according to the behavior data is specifically as follows: Track the customer's movement path in the store through motion target detection technology, and analyze the customer's points of interest based on the stay time and location information; If the stay time is more than 10 seconds but less than 20 seconds, the point of interest is determined to be a primary point of interest; If the stay time is more than 20 seconds but less than 30 seconds, the point of interest is determined to be an intermediate point of interest; If the stay time exceeds 30 seconds, the point of interest is determined to be a high-level point of interest.

8. The method according to claim 1, characterized in that The preliminary recommendation list is generated by a preset recommendation strategy, and the recommendation strategy assigns different weights to the behavior data and the customer attributes, specifically: The recommendation strategy adopts a linear weighted method to obtain the types of goods required by the customer based on the behavior data, and determines the types of goods that meet the customer profile in combination with the customer attributes to obtain a preliminary recommendation list; The weight of the behavior data in the recommendation list is greater than the weight of the customer attributes.

9. The method according to claim 4, characterized in that Also includes: assigning corresponding weights to the facial features and the clothing features based on the recommendation strategy, wherein the weight of the facial features is greater than the weight of the clothing features; assigning corresponding weights to the age range and the gender based on the recommendation strategy, wherein the weight of the gender in the facial features is greater than the weight of the age range; Based on the recommendation strategy, corresponding weights are assigned to the color attributes and the clothing style, and the weight of the clothing style in the clothing features is greater than the weight of the color attributes.

10. The method according to claim 7, characterized in that Also includes: Based on the recommendation strategy, each of the interest points is assigned a corresponding weight, the weight of the high-level interest point in the behavior data is greater than the weight of the intermediate interest point, and the weight of the intermediate interest point in the behavior data is greater than the weight of the low-level interest point.

Images