Clothing display effect evaluation optimization method and system based on visual analysis

Abstract

The application discloses a clothing display effect evaluation optimization method and system based on visual analysis, relates to the technical field of computer vision, and comprises the following steps: acquiring a continuous video frame stream, depth point cloud data and texture feature data of a display area; extracting an interactive behavior track of a customer and a clothing single product, and calculating a space disturbance index; constructing a dynamic interactive topology graph, and calculating a graph condensation coefficient; calculating a time-varying slope of the space disturbance index and a time-varying slope of the graph condensation coefficient; determining a display state based on the time-varying slope of the space disturbance index and a display conversion potential index, and outputting a corresponding maintenance strategy; when the time-varying slope of the space disturbance index is greater than or equal to zero, the display conversion potential index is positively correlated with the time-varying slope of the graph condensation coefficient and negatively correlated with the time-varying slope of the space disturbance index; and the application has the beneficial effects that the defects of the traditional scheme that cannot identify the cause of disordered display and the rigid alarm mechanism are solved, and the differentiated triggering of the display maintenance strategy is realized.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a method and system for evaluating and optimizing the effect of clothing display based on visual analysis. Background Technology

[0002] With the digital transformation of the offline retail industry, the refined operation of apparel stores has become increasingly important. Among various display methods, stacked displays are widely used because they can effectively improve space efficiency. However, stacked displays are prone to becoming cluttered during customer browsing and selection, which not only affects the brand's visual image but may also reduce subsequent customers' willingness to purchase. Therefore, introducing apparel display evaluation technology based on visual analysis, which uses visual sensors to monitor and quantify the status of display areas in real time, and guides store staff to perform timely stocking and maintenance, has significant commercial application value for reducing store manual inspection costs, improving the customer shopping experience, and promoting terminal sales.

[0003] Currently, existing visual assessment solutions for apparel displays typically employ static comparison algorithms based on absolute image features (such as pixel difference, edge extraction, or texture complexity) to calculate the physical clutter level of the display. When the detected clutter level exceeds a single preset threshold, the system triggers a merchandising alarm. However, this conventional technology can only identify "superficial clutter" in practical applications, failing to accurately determine "whether merchandising is truly necessary," and cannot distinguish between "reasonable clutter caused by customers' in-depth shopping and styling" and "malicious clutter caused by worthless discarding." This rigid "threshold-only" alarm mechanism not only easily leads to ineffective order dispatch and unnecessary consumption of store manpower, but also forcibly interrupts consumers' immersive shopping process during peak hours by sending staff, ultimately failing to optimize the display and negatively impacting the store's actual commercial conversion.

[0004] Therefore, a method and system for evaluating and optimizing the effect of clothing display based on visual analysis is proposed. Summary of the Invention

[0005] In view of the above-mentioned state of the prior art, this application is hereby filed. Embodiments of this application provide a method and system for evaluating and optimizing the effect of clothing displays based on visual analysis. This system can address the core shortcomings of existing technologies, such as the inability to distinguish the causes of disordered displays and the rigidity of single-threshold alarm mechanisms, thereby achieving refined and scenario-based dynamic evaluation of stacked clothing display status.

[0006] According to one aspect of this application, a method for evaluating and optimizing clothing display effects based on visual analysis is provided, applied to the optimization of stacked clothing display areas, comprising: acquiring visual temporal data of the display area, the visual temporal data including continuous video frame streams, depth point cloud data, and texture feature data; extracting the interaction behavior trajectories between customers and individual clothing items in the display area based on the visual temporal data, and calculating a spatial disturbance index characterizing the intensity of physical disturbance generated in the display area under customer interaction; constructing a dynamic interaction topology map characterizing the spatiotemporal correlation strength between different clothing items based on the interaction behavior trajectories, and calculating a graph cohesion coefficient characterizing the degree of node association cohesion in the dynamic interaction topology map; and calculating the time-varying slope of the spatial disturbance index within a current preset step period. and the time-varying slope of the graph's cohesion coefficient. ; Determine the If the value is less than zero, then the current display state of the display area is determined to be in a disturbance attenuation state; otherwise: based on a preset nonlinear mapping function, the value is determined to be less than zero. With the Coupled calculations are performed to obtain the display conversion potential energy index; based on the numerical range of the display conversion potential energy index, the display state is determined to be either a structured interaction state or an unstructured interaction state; a corresponding display maintenance strategy is generated and output according to the display state; wherein, in Within the constraint range, the display conversion potential energy index and Positively correlated with It shows a negative correlation.

[0007] According to another aspect of this application, a visual analysis-based clothing display effect evaluation and optimization system is provided, applied to the display optimization of stacked clothing display areas. The system is characterized by comprising: a data acquisition module for acquiring visual temporal data of the display area, the visual temporal data including continuous video frame streams, depth point cloud data, and texture feature data; a data processing module for extracting the interaction behavior trajectories between customers and individual clothing items in the display area based on the visual temporal data, and calculating a spatial disturbance index characterizing the intensity of physical disturbance generated in the display area under customer interaction; a graph construction processing module for constructing a dynamic interaction topology graph characterizing the spatiotemporal correlation strength between different clothing items based on the interaction behavior trajectories, and calculating a graph cohesion coefficient characterizing the degree of node association cohesion in the dynamic interaction topology graph; and a time-varying slope calculation module for calculating the time-varying slope of the spatial disturbance index within a current preset step period. and the time-varying slope of the graph's cohesion coefficient. The status determination module is used to determine the status of the... If the value is less than zero, then the current display state of the display area is determined to be in a disturbance attenuation state; otherwise: based on a preset nonlinear mapping function, the value is determined to be less than zero. With the A coupled calculation is performed to obtain the display conversion potential energy index; based on the numerical range of the display conversion potential energy index, the display state is determined to be either a structured interaction state or an unstructured interaction state; a strategy generation module is used to generate and output corresponding display maintenance strategies according to the display state; wherein, in Within the constraint range, the display conversion potential energy index and Positively correlated with It shows a negative correlation.

[0008] According to another aspect of this application, an electronic device is provided, including a memory and a processor, the memory being used to store computer-executable instructions, and the processor being used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method described above.

[0009] According to another aspect of this application, a computer storage medium is provided that stores computer-executable instructions thereon, which, when executed by a processor, implement the steps of the method described above.

[0010] Compared with existing technologies, the visual analysis-based clothing display effect evaluation and optimization method and system according to the embodiments of this application can realize the structured quantification of customer purchasing intentions through dynamic interactive topology maps, capture the dynamic changing trend of display disturbances and customer intentions through time-varying slopes, and realize the three-state fine differentiation of display status through non-linearly coupled display transformation potential energy indicators. This solves the defects of traditional solutions that cannot identify the causes of display disorder and have rigid alarm mechanisms, and realizes differentiated and intelligent triggering of display maintenance strategies. Attached Figure Description

[0011] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0012] Figure 1 This is a flowchart of the method for evaluating and optimizing the effect of clothing display based on visual analysis according to the present invention.

[0013] Figure 2 This is a block diagram of the visual analysis-based clothing display effect evaluation and optimization system of the present invention.

[0014] Figure 3 This is a block diagram of an electronic device. Detailed Implementation

[0015] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0016] Exemplary method:

[0017] Figure 1 The illustration shows a method for evaluating and optimizing the effect of clothing display based on visual analysis according to an embodiment of this application, including steps S1 to S6:

[0018] like Figure 1 As shown, in step S1, visual temporal data of the display area is acquired. The visual temporal data includes continuous video frame streams, depth point cloud data, and texture feature data.

[0019] Specifically, an RGB-D depth camera is deployed above the display area as a hardware acquisition device to simultaneously acquire visual temporal data, including continuous video frame streams, depth point cloud data, and texture feature data. The continuous video frame streams are used to capture customer human interaction processes, the depth point cloud data is used to capture the physical three-dimensional deformation state of the clothing display, and the texture feature data is used to capture surface misalignment information of the clothing display. The simultaneous acquisition of these three data streams provides a multi-dimensional perceptual foundation for subsequent accurate quantification of the degree of physical disturbance in the display.

[0020] For example, suppose a fast-fashion store's women's outerwear display case shows three items: A plaid coat, B a solid-color trench coat, and C a down jacket. Specifically, an RGB-D depth camera is deployed above this display case, continuously and synchronously acquiring three channels of visual temporal data within a preset step period T: a continuous video frame stream records the entire process of a customer entering the display area, touching and examining the clothing; depth point cloud data captures the three-dimensional deformation caused by the clothing being moved in real time; and texture feature data records the misalignment of the clothing surface. These three channels of data will serve as the input basis for subsequent steps.

[0021] return Figure 1 In step S2, based on visual time-series data, the interaction behavior trajectory between customers and clothing items in the display area is extracted, and the spatial disturbance index, which characterizes the intensity of physical disturbance generated in the display area under customer interaction, is calculated.

[0022] Specifically, step S2 requires the simultaneous extraction of two types of key information from multidimensional visual time-series data: first, the interaction behavior trajectory between customers and individual clothing items, used to construct a dynamic interaction topology map in subsequent step S3; and second, spatial disturbance indicators that quantify the degree of physical damage to the display area. The quantification method for the spatial disturbance indicator is one of the core innovations of this application—unlike traditional methods based on image difference or simple threshold judgment, this application uses relative entropy (KL divergence) to measure the degree of deviation of the display state from the baseline state, thereby accurately quantifying the true physical disturbance of the display area from an information theory perspective and reducing the interference of the complexity of the clothing's patterns on disturbance detection.

[0023] Specifically, the calculation of the spatial disturbance index includes: dividing the display area into several sub-spatial units; extracting the height variance of the depth point cloud and the distribution complexity of the texture gradient within each sub-spatial unit based on depth point cloud data and texture feature data; performing weighted fusion calculation on the height variance and distribution complexity based on a preset mapping function to obtain a state value used to characterize the physical deformation characteristics of a single sub-spatial unit; statistically analyzing the distribution probability of the state values ​​of several sub-spatial units within a preset value range as the current distribution; and calculating the relative entropy of the current distribution relative to a preset reference distribution as the spatial disturbance index.

[0024] The aforementioned preset mapping function adopts a weighted linear combination. Specifically, the state value is equal to the product of the depth point cloud height variance and the preset first weight coefficient, plus the product of the texture gradient distribution complexity and the preset second weight coefficient. Both weight coefficients are positive values ​​and can be preset and adjusted according to the differences in the contribution of physical deformation and texture misalignment to the perception of disturbance in different display scenarios, so as to adapt to clothing categories with different materials and display methods.

[0025] In the above steps, the formula for calculating relative entropy is:

[0026] ;

[0027] in, This is the reference distribution probability of the state values ​​of each subspace unit collected under a preset physical reference state. This represents the current distribution probability of the state values ​​of each subspace unit at the current moment. This refers to the range of values ​​to which the state value belongs. The core advantage of this design is that if the garment itself has a complex pattern but has not been moved, then... and Maintaining consistency, with a relative entropy of 0 and a spatial disturbance index of zero, solves the interference problem caused by the inherent patterns of clothing; conversely, if physical disturbance occurs in the display, Will deviate The relative entropy increases accordingly, accurately reflecting the degree of disturbance in the display.

[0028] To facilitate understanding of the above calculation process, we will continue to use the previous example of the outerwear display stand. (Step cycle) Before starting, a reference distribution was collected when the display stand was in a clean and tidy condition. The display stand is divided into several sub-spatial units. At this point, the height variance of each unit is minimal, the texture gradient distribution is stable, and the state values ​​are concentrated in the low-disturbance range. It exhibits a clear pattern of concentrated low-disturbance distribution. During the period... Inside, if a customer only looked at and restored style A of the plaid coat, then the current distribution... and Almost identical, A value close to 0 indicates minimal disturbance to the display; if the customer has significantly moved all three garments and not returned them to their original positions, then... Accumulation is highly probable in areas of high disturbance. The spatial disturbance index increases significantly, accurately reflecting the degree of disorder in the display. In particular, even though the plaid coat A has a complex pattern, its texture feature data distribution remains consistent with the baseline state when it is not moved, with a relative entropy contribution of zero, thus reducing pattern interference.

[0029] It should be noted that the above relative entropy calculation formula has a boundary problem of "denominator being zero" in engineering implementation: when a certain value interval is within the reference distribution... The probability of it being in the distribution is zero, while the current distribution... When a non-zero probability appears in this interval, The term will tend towards negative infinity, leading to computational overflow. This situation is not uncommon in real-world scenarios—for example, after the display stand is significantly modified, extreme deformation ranges that were never observed under the baseline state appear. In this case, directly applying the original formula will render the entire disturbance index invalid.

[0030] Therefore, in engineering implementation, a reference distribution can be used. and current distribution A unified Laplace smoothing is introduced: a minimal constant is superimposed on the probability estimate for each interval of values. (like = 1× This ensures that the probability value in any interval is strictly greater than zero, thus guaranteeing the validity of the logarithmic term calculation. The smoothed probability estimation formula is: ,in The current distribution is within the range of values. Sample count within, The total number of samples, This represents the total number of values ​​within the range. The smoothing process is the same. After smoothing, The calculation results remain bounded under any arrangement state, thus ensuring the stability of the spatial disturbance index under extreme scenarios.

[0031] return Figure 1 In step S3, a dynamic interactive topology map representing the spatiotemporal correlation strength between different clothing items is constructed based on the interactive behavior trajectory, and the graph cohesion coefficient representing the degree of node association cohesion of the dynamic interactive topology map is calculated.

[0032] Specifically, step S3, based on the interactive behavior trajectory extracted in step S2, expresses the customer's shopping behavior patterns in the form of a graph structure, thereby realizing the transformation from discrete behavior trajectory data to a structured intent network. The construction of a dynamic interactive topology graph is another core innovation of this application—by dynamically weighting the edges of the graph by integrating the customer's spatial dwell time and the rate of change of movement at the trajectory endpoints, the graph can reflect the spatiotemporal correlation strength between different clothing items in real time. Furthermore, the graph cohesion coefficient quantifies the degree of customer intent aggregation, providing crucial intent dimension information for subsequent display status determination.

[0033] Specifically, the construction of the dynamic interactive topology graph includes the following steps: extracting the temporal switching trajectory of a customer touching different clothing items within a preset time window based on the interactive behavior trajectory; constructing an initial topology graph, where nodes represent clothing items touched by the customer and edges represent corresponding temporal switching trajectories; determining the intensity of the interactive action based on the spatial dwell time of the temporal switching trajectory within the corresponding clothing item area and the rate of change of motion of the endpoints of the temporal switching trajectory; and assigning real-time weights to the corresponding edges of the initial topology graph based on the intensity of the interactive action to obtain the dynamic interactive topology graph.

[0034] The graph cohesion coefficient is calculated based on the edge weights of the dynamic interactive topology graph. Specifically, the graph cohesion coefficient is equal to the sum of the weights of all edges in the graph, divided by half the product of the number of nodes and the number of nodes minus one. That is, the normalized weighted edge density is used as an overall measure of the degree of customer intent aggregation. When only a few edges in the graph have high weights, the graph cohesion coefficient is low, indicating that customer intent is dispersed; when multiple edges have high weights simultaneously, the graph cohesion coefficient increases, indicating that customers have generated obvious association intentions among multiple clothing item combinations, and the overall aggregation degree is high.

[0035] In the construction of the aforementioned dynamic interactive topology graph, the intensity of interactive actions is determined by two dimensions: spatial dwell time reflects the time a customer spends on a particular garment, with longer dwell times indicating higher levels of attention; and the rate of change of motion at the trajectory endpoints characterizes the certainty of the customer's intent during switching actions, with lower rates of change indicating more relaxed switching actions and clearer intent. After merging these two dimensions, the edge weights can dynamically and realistically reflect the intensity of the customer's associated intent with different combinations of garment items, thus enabling the graph cohesion coefficient to comprehensively depict the aggregation and dispersion of customer intent within the current display area. Specifically, the intensity of interactive actions is calculated using a weighted product: the product of spatial dwell time and a preset dwell weight coefficient is multiplied by the inverse of the rate of change of motion (or the complement after normalization) and then multiplied by a preset rate of change weight coefficient to obtain the interactive action intensity value; this value is then normalized to a preset weight range and used as the weight of the corresponding edge. The above design results in higher edge weights when the dwell time is longer and the switching action is more relaxed (lower rate of change of movement), thus reflecting the strength of the customer's active association intention with the product combination.

[0036] Taking the aforementioned coat display stand as an example, the above construction process can be understood more intuitively. Within period T, the behavioral trajectories of two customers are extracted from the continuous video frame stream: Customer A's temporal switching trajectory is A→B→A (first touching the plaid coat, then touching the solid-color trench coat, then returning to the plaid coat), lingering on item A for a total of 18 seconds and on item B for 8 seconds, with a slow switching action (low rate of motion change); Customer B's temporal switching trajectory is C→B (first touching the down jacket, then touching the solid-color trench coat), lingering on item C for 5 seconds and on item B for 3 seconds, with a faster switching action (higher rate of motion change). Based on this, an initial topology graph is constructed, with nodes {A, B, C} and edges {AB, CB}. Weighting is based on the intensity of the interactive actions: Customer A's A→B switching action is leisurely and has a long dwell time, corresponding to edge AB receiving a higher weight (e.g., 0.72); Customer B's C→B switching action is fast and has a short dwell time, corresponding to edge CB receiving a lower weight (e.g., 0.31). In the resulting dynamic interactive topology graph, the strong correlation between edges AB indicates that customers have a clear intention to actively explore the combination of "plaid coat + solid color trench coat". The graph cohesion coefficient is high, and the degree of intention aggregation is significant.

[0037] It should be noted that in scenarios with dense interaction, the construction of the aforementioned topology graph faces the problem of "edge breakage": when multiple customers simultaneously touch the display area, mutual occlusion between their hands may cause some trajectory points in the video frame stream to temporarily disappear, resulting in breakpoints in the time-series switching trajectory between consecutive frames. This leads to a sharp drop or even zeroing of the weights of the corresponding edges in the dynamic interaction topology graph, causing abnormal fluctuations in the graph cohesion coefficient and interfering with subsequent display status determination. Therefore, in practice, a Kalman filter can be introduced to interpolate and complete the trajectory disappearance sections. Specifically, within a preset sliding observation window, the motion state is continuously modeled based on the historical position and velocity vector of the trajectory endpoints; when a trajectory breakpoint appears in the continuous video frame stream, the missing trajectory points are predictively interpolated within this time window using the motion equation, so that the time-series switching trajectory can be made continuous; only when the duration of the breakpoint exceeds the preset motion occlusion limit is it determined as a true termination of interaction and the edge weights are updated accordingly. The Kalman filter interpolation mechanism described above can ensure the continuity of edge weights in dynamic interactive topology graphs under dense interaction scenarios, thereby improving the robustness of graph cohesion coefficient calculation.

[0038] return Figure 1 In step S4, the time-varying slope of the spatial disturbance index is calculated within the current preset step size period. and the time-varying slope of the graph cohesion coefficient. .

[0039] After calculating the spatial perturbation index and the graph cohesion coefficient, step S4 further transforms their instantaneous states into time-series features with trend prediction significance by extracting the time-varying slope, a dynamic dimension index. Time-varying slope This reflects the changing trend of the degree of physical disturbance to the display: A value less than 0 indicates that the disturbance is decaying, meaning that customers have stopped interacting with the display. A value ≥0 indicates that the disturbance is ongoing or increasing, and the display is still in an active interactive phase. Time-varying slope This reflects the changing trend in the degree of customer intent concentration, and... Together, they constitute the input features for the nonlinear coupling determination in step S5. The combination of the two slopes thus achieves a refined dynamic distinction of the display state.

[0040] Taking the aforementioned outerwear display stand as an example, in the cycle Inside, as customer A continued to browse through styles A and B, the spatial disturbance index steadily increased in the first half of the cycle. >0; Meanwhile, the graph cohesion coefficient increases as the weight of edge AB increases. A value greater than 0 indicates that customer intent is rapidly converging towards combination A and B. Entering the latter half of the cycle, customer A stops browsing and leaves, and the disturbance index gradually declines. If the value is less than 0, the process proceeds to determine the disturbance decay state. In another scenario, if the customer is within a certain period... If three pieces of clothing are continuously searched inside and no signs of gathering are observed, then... >0 and If the slope is ≈0 or even negative, the difference between the two slopes will trigger the determination of the unstructured interaction state in the next step.

[0041] It should be noted that the above time-varying slope and The time-varying slope is calculated by performing linear least squares fitting on the time-series sampled data within the current preset step size period: with each sampling time as the independent variable and the corresponding spatial perturbation index or graph cohesion coefficient as the dependent variable, the slope of the fitted linear regression line is the corresponding time-varying slope. When the number of sampling points is insufficient to support effective fitting (e.g., the number of effective frames within the period is lower than the preset minimum frame threshold), the time-varying slope is set to zero to avoid unstable slope estimation due to insufficient samples.

[0042] return Figure 1 In step S5, the determination is made. If the value is less than zero, then the current display area is determined to be in a disturbance attenuation state; otherwise: based on a preset nonlinear mapping function... and Coupled calculations are performed to obtain the display conversion potential energy index; based on the numerical range of the display conversion potential energy index, the display state is determined to be either a structured interaction state or an unstructured interaction state; among which, in Within the constrained range, the display conversion potential energy index and Positively correlated with It shows a negative correlation.

[0043] The formula for calculating the display conversion potential energy index is:

[0044] ;

[0045] in, The preset intention gain coefficient, The preset disturbance attenuation penalty coefficient, It is a natural constant. This is an indicator of the potential energy of the display transformation.

[0046] The above formula integrates the spatial disturbance dimension and the customer intent dimension into a unified display conversion potential index. This allows for the differentiation of the three states of the display. Compared to the traditional coarse-grained decision-making method that relies solely on a single disturbance signal to trigger merchandising, this application uses a potential energy index to distinguish between "whether the display is chaotic" and "whether the chaos stems from high-intention shopping behavior," thereby avoiding two types of decision-making errors: unnecessarily interrupting sales opportunities during peak shopping periods and delaying maintenance opportunities due to ineffective disturbances.

[0047] The above display conversion potential indicators The nonlinear coupling design exhibits the following key characteristics: Within the constraint interval ≥ 0, and A positive correlation means that the more concentrated the customer intent (the faster the graph cohesion coefficient increases), the higher the potential energy, and the more likely it is to be judged as a structured interaction state (high-intent purchase selection); at the same time, and A negative correlation means that the faster the degree of display disturbance increases, the lower the potential energy, reflecting that a high disturbance rate is often accompanied by non-purposeful searching behavior. (Exponential term) As a penalty factor, even at high perturbation rates, The larger the potential energy, the less likely it is to be misjudged as high-intention shopping. and The initial values ​​for the two coefficients can be determined by referring to the following empirical range: ∈[0.5, 2.0], ∈[1.0, 4.0], the specific value is continuously calibrated after actual deployment using a data-driven adaptive tuning method. The tuning mechanism uses the average order value associated with the transaction as a business monitoring signal: if the associated rate increases in the corresponding period after implementing the inventory suppression strategy, the restrictions are adaptively relaxed. and Threshold tolerance (i.e., increase) , reduce This allows for the protection of customer purchasing behavior within a broader potential energy range; conversely, if the linkage rate does not improve, the threshold is tightened to enhance the sensitivity of immediate maintenance triggers, thereby achieving a closed-loop self-evolution of the business decision-making model.

[0048] It should be further explained that the actual effect of the above tuning direction is as follows: Increase Overall improvement The amplitude of the same At a horizontal level, potential energy more easily reaches the high potential energy conversion threshold, thus expanding the range for determining structured interaction states; reducing This weakens the effect of the exponential penalty factor on high The increased suppression intensity of the scenario reduces the potential energy decay when the disturbance rate is high, further lowering the probability of misjudging high-intent purchasing behavior. These two adjustments work synergistically, gradually converging towards a more lenient judgment envelope under the positive feedback of improved linkage rate, ensuring that the tuning direction remains consistent with business objectives.

[0049] Using the data from the aforementioned outerwear display stand in two different scenarios as examples, this illustrates... The calculation and judgment process. Scenario 1: Customer A calmly and continuously flips through items A and B, periodically... Inside =0.15 (the disturbance continues to increase but at a moderate rate). =0.42 (The graph shows a rapid increase in the cohesion coefficient, indicating significant intentional aggregation), substituting into the formula. ,set up =1.0, =2.0, then =(1.0×0.42)× ≈0.31, assuming the high potential energy conversion threshold is 0.25, then A value ≥0.25 indicates a structured interaction state, suggesting the customer is engaging in high-intent combination shopping and should be discouraged from restocking. Scenario 2: The customer quickly flips through three garments without any tendency to cluster together. =0.35 (rapid accumulation of disturbances) =0.02 (the cohesion coefficient of the graph remains almost unchanged), then =(1.0×0.02)× If the value is approximately 0.01, and the low potential energy conversion threshold is 0.08 and the absolute value of the spatial disturbance index exceeds the disturbance threshold, then it is determined to be an unstructured interaction state, indicating that the display has been invalidally searched and damaged, and maintenance should be triggered immediately.

[0050] The determination of whether the display status is a structured or unstructured interactive state includes the following rules: determine whether the display conversion potential energy index is greater than or equal to the preset high potential energy conversion threshold; if so, the display status is determined to be a structured interactive state. Determine whether the display conversion potential energy index is less than the preset low potential energy conversion threshold, and whether the absolute value of the spatial disturbance index within the current preset step size period is greater than the preset disturbance threshold; if so, the display status is determined to be an unstructured interactive state.

[0051] It should be noted that the three-state determination mechanism mentioned above (disturbance decay state, structured interaction state, and unstructured interaction state) is not a simple linear hierarchy, but rather based on... After initial segmentation of the symbols, then... Utilization of intervals ≥0 The numerical ranges are finely distinguished. Specifically, structured interaction states require... A value not lower than the high potential energy conversion threshold indicates that customer intent is concentrated and disturbances are controlled; the sales environment should be protected and inventory management suppressed. Unstructured interaction states, on the other hand, require… If the value is below the low potential energy conversion threshold and the absolute value of the spatial disturbance index exceeds the disturbance threshold, both conditions confirm the "high disturbance + low intent" characteristic, thus triggering immediate maintenance. Within the interval between the two thresholds, no maintenance action is triggered. This method maintains the current cycle's display status assessment result unchanged within this interval and re-evaluates the timing features after the next preset step size cycle, waiting for the potential energy index to converge to a clear interval. This demonstrates the method's fault-tolerant handling of fuzzy boundaries. After completing the display status determination, step S6 is entered, where the corresponding maintenance instruction is generated based on the determination result.

[0052] return Figure 1 In step S6, a corresponding display maintenance strategy is generated and output based on the display status.

[0053] Specifically, step S6 implements differentiated closed-loop maintenance strategies based on the three display states output in step S5, transforming the evaluation results into business instructions that can be directly implemented, thereby achieving a complete closed loop from intelligent perception to business execution.

[0054] The maintenance strategies for each display state are as follows: If the display state is a structured interaction state, a merchandising suppression strategy is output, which is used to indicate that merchandising in the current display area should be stopped; if it is determined to be an unstructured interaction state, an immediate maintenance strategy is output, which is used to indicate that merchandising should be performed on clothing items whose edge weights in the dynamic interaction topology graph are lower than a preset lower limit; if it is determined to be a disturbance decay state, a merchandising suspension strategy is output, which is used to indicate that merchandising in the current display area should be stopped, and when the absolute value of the spatial disturbance index is lower than a preset steady-state threshold for a preset number of consecutive frames, a reference distribution update is triggered, and the preset reference distribution is re-collected and replaced with the current visual time-series data, thereby avoiding reference distribution contamination caused by customers returning after a short period of absence.

[0055] The differentiated design of the three display maintenance strategies mentioned above reflects the core decision-making logic of this method: the merchandising suppression strategy (structured interaction state) protects the shopping environment for high-intent customers and prevents merchandising from interrupting the shopping process; the instant maintenance strategy (unstructured interaction state) performs precise targeted merchandising for clothing items with edge weights below a preset lower limit in the dynamic interaction topology graph, concentrating maintenance resources on items that are truly in an ineffective disturbance state, rather than performing global reorganization of the entire display area; and the merchandising suspension strategy (disturbance decay state) updates the reference distribution. Adaptive benchmark correction is achieved, ensuring that subsequent spatial perturbation calculations always use the latest steady-state arrangement as a reference, thereby enabling the entire evaluation method to have the ability to continuously self-calibrate.

[0056] Taking the aforementioned outerwear display stand as an example, the differentiated execution process of the three strategies can be intuitively understood. In scenario one (structured interaction state), the product sorting suppression strategy is output, instructing store staff not to tidy up the outerwear display stand, protecting customer A's ongoing purchase of a combination of styles A and B, and avoiding disturbance. In scenario two (unstructured interaction state), the immediate maintenance strategy is output. Since the weight of the CB side (0.31) is lower than the weight of the AB side (0.72) and lower than the preset lower limit, it is instructed to prioritize the sorting of style C down jackets, while styles A and B do not require immediate intervention because they are at the customer's center of attention. After all customers leave, the disturbance index decreases accordingly. If the value is less than 0, output the merchandising suspension policy. While instructing to pause merchandising, update the reference distribution with the current cleaned-up display state. This will establish a new benchmark for disturbance detection in the next cycle.

[0057] In addition to the three maintenance strategies mentioned above, this application also provides an optimization suggestion mechanism for the display layout itself. Specifically, the method of this application also includes: extracting target clothing item combinations whose edge weights exceed a preset strong association threshold in the dynamic interactive topology graph; calculating the physical spatial distance between each clothing item in the target clothing item combination within the current display area; if the physical spatial distance exceeds a preset visual linkage distance, generating a layout reconstruction suggestion, which is used to indicate that the target clothing item combination should be displayed adjacent to each other in physical space.

[0058] The aforementioned layout reconstruction suggestions serve as an advanced optimization function of this method. Based on the strong correlation information in the dynamic interactive topology map, customer behavior data is transformed into a decision-making basis for display layout optimization. When strongly correlated clothing item combinations are physically too far apart (exceeding the preset visual linkage distance), customers need to move back and forth between different display areas to complete the combination purchase, which not only increases the shopping path cost but also leads to cross-area weak linkage noise signals in the topology map. By displaying strongly correlated item combinations adjacent to each other, the shopping experience can be effectively improved, while the structure of the dynamic interactive topology map becomes clearer, further improving the accuracy of subsequent display status determination.

[0059] Continuing with the aforementioned coat display example, after accumulating data over several periods, it was found that the edge weight between style A (plaid coat) and style B (solid-color trench coat) consistently exceeded the strong association threshold (e.g., 0.65), indicating that customers repeatedly associated and selected these two items. However, physical measurements showed that the distance between styles A and B on the display stand was 1.2 meters, exceeding the preset visual linkage distance (e.g., 0.8 meters). Therefore, a layout restructuring suggestion was generated, instructing that style B (solid-color trench coat) be moved to a position adjacent to style A (plaid coat), allowing customers to simultaneously access both strongly associated items without moving, thereby reducing shopping path costs and increasing the conversion rate of set sales.

[0060] In addition, to avoid misjudgments by the system due to customers briefly organizing merchandise or continuously browsing, the above solution can also consider incorporating the intensity of customer interaction actions when determining the display status. When the display conversion potential index shows a downward trend but the intensity of customer interaction actions remains high, the system maintains the current reference distribution without updating; when the display conversion potential index continues to decline and the intensity of customer interaction actions falls below a preset threshold, it is determined to be a disturbance decay state and the reference distribution is updated. When determining the merchandising trigger, a comprehensive judgment is made by combining the display conversion potential index and the intensity of customer interaction actions. When the display conversion potential index is higher than the preset threshold and the intensity of customer interaction actions is at a low level, the current merchandise display is determined to be in a state of being unvisited but disorganized, thereby triggering a merchandising prompt.

[0061] In summary, the visual analysis-based method for evaluating and optimizing apparel display effects proposed in this application achieves refined, dynamic, and adaptive evaluation of apparel display status through multi-dimensional visual temporal data acquisition, spatial perturbation quantification based on relative entropy, intent network representation based on dynamic interactive topology graphs, and three-state display status recognition based on nonlinear coupling judgment. The evaluation results are directly transformed into differentiated display maintenance strategies, forming a complete closed loop from perception to decision-making. The core innovation of this method lies in: using relative entropy to solve the interference problem of inherent patterns on perturbation detection in traditional methods; using dynamic weighted topology graphs to transform discrete customer behavior trajectories into structured intent representations; and using nonlinear potential energy functions to cross-domain couple the physical perturbation dimension and the customer intent dimension, achieving a more accurate display status recognition capability compared to traditional single-dimensional judgments. This provides a systematic technical solution for intelligent display management in retail scenarios.

[0062] Exemplary system:

[0063] Figure 2The illustration shows a visual analysis-based clothing display effect evaluation and optimization system according to an embodiment of this application, including: a data acquisition module for acquiring visual temporal data of the display area, the visual temporal data including continuous video frame streams, depth point cloud data, and texture feature data; a data processing module for extracting the interaction behavior trajectory between customers and clothing items in the display area based on the visual temporal data, and calculating a spatial disturbance index representing the intensity of physical disturbance generated in the display area under customer interaction; a graph construction processing module for constructing a dynamic interaction topology graph representing the spatiotemporal correlation strength between different clothing items based on the interaction behavior trajectory, and calculating a graph cohesion coefficient representing the degree of node association cohesion in the dynamic interaction topology graph; and a time-varying slope calculation module for calculating the time-varying slope of the spatial disturbance index within the current preset step size period. and the time-varying slope of the graph cohesion coefficient. The status determination module is used to determine... If the value is less than zero, then the current display area is determined to be in a disturbance attenuation state; otherwise: based on a preset nonlinear mapping function... and Coupled calculations are performed to obtain the display conversion potential energy index; based on the numerical range of the display conversion potential energy index, the display state is determined to be either a structured interaction state or an unstructured interaction state; the strategy generation module is used to generate and output corresponding display maintenance strategies according to the display state; among them, in Within the constrained range, the display conversion potential energy index and Positively correlated with It shows a negative correlation.

[0064] In one example, the calculation of the spatial disturbance index includes: dividing the display area into several sub-spatial units; extracting the height variance of the depth point cloud and the distribution complexity of the texture gradient within each sub-spatial unit based on depth point cloud data and texture feature data; performing weighted fusion calculation on the height variance and distribution complexity based on a preset mapping function to obtain a state value used to characterize the physical deformation characteristics of a single sub-spatial unit; statistically analyzing the distribution probability of the state values ​​of several sub-spatial units within a preset value range as the current distribution; and calculating the relative entropy of the current distribution relative to a preset reference distribution as the spatial disturbance index.

[0065] In one example, the construction of a dynamic interaction topology graph includes: extracting the temporal switching trajectory of a customer touching different clothing items within a preset time window based on the interaction behavior trajectory; constructing an initial topology graph, where nodes represent clothing items touched by the customer and edges represent corresponding temporal switching trajectories; determining the intensity of the interaction action based on the spatial dwell time of the temporal switching trajectory within the corresponding clothing item's area and the rate of change of motion of the endpoints of the temporal switching trajectory; and assigning real-time weights to the corresponding edges of the initial topology graph based on the intensity of the interaction action to obtain the dynamic interaction topology graph.

[0066] In one example, the formula for calculating the display conversion potential index is: ;in, The preset intention gain coefficient, The preset disturbance attenuation penalty coefficient, It is a natural constant. This is an indicator of the potential energy of the display transformation.

[0067] In one example, determining whether the display state is a structured interaction state or an unstructured interaction state includes: determining whether the display conversion potential energy index is greater than or equal to a preset high potential energy conversion threshold; if so, the display state is determined to be a structured interaction state; determining whether the display conversion potential energy index is less than a preset low potential energy conversion threshold, and whether the absolute value of the spatial disturbance index within the current preset step size period is greater than a preset disturbance threshold; if so, the display state is determined to be an unstructured interaction state.

[0068] In one example, a corresponding display maintenance strategy is generated and output based on the display status, including: if the display status is a structured interaction status, a merchandising suppression strategy is output, which is used to indicate that merchandising in the current display area should be stopped; if it is determined to be an unstructured interaction status, an immediate maintenance strategy is output, which is used to indicate that merchandising should be performed on clothing items whose edge weights in the dynamic interaction topology graph are lower than a preset lower limit; if it is determined to be a disturbance decay status, a merchandising suspension strategy is output, which is used to indicate that merchandising in the current display area should be stopped, and the preset reference distribution is updated according to the current visual time series data.

[0069] In one example, the strategy generation module further includes: extracting target clothing item combinations whose edge weights exceed a preset strong association threshold in the dynamic interactive topology graph; calculating the physical spatial distance between each clothing item in the target clothing item combination within the current display area; and generating layout reconstruction suggestions if the physical spatial distance exceeds a preset visual linkage distance. The layout reconstruction suggestions are used to instruct the target clothing item combination to be displayed adjacent to each other in physical space.

[0070] Exemplary electronic device:

[0071] Figure 3 A block diagram of an electronic device according to an embodiment of this application is illustrated.

[0072] like Figure 3 As shown, the electronic device includes one or more processors and memory.

[0073] A processor can be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and can control other components in an electronic device to perform desired functions.

[0074] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0075] In one example, the electronic device may also include input devices and output devices, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0076] Of course, for the sake of simplicity, Figure 3 Only some of the components of the electronic device relevant to this application are shown in this illustration; components such as buses, input / output interfaces, etc., are omitted. In addition, the electronic device may include any other suitable components depending on the specific application.

[0077] Exemplary computer-readable medium:

[0078] Embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps described in the "Exemplary Methods" section above according to the various embodiments of this application.

[0079] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0080] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not restrict the application from being implemented using the specific details described above.

[0081] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0082] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0083] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0084] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for evaluating and optimizing clothing display effects based on visual analysis, applied to the optimization of stacked clothing display areas, characterized in that: include: Acquire visual temporal data of the display area, the visual temporal data including continuous video frame streams, depth point cloud data and texture feature data; Based on the visual time-series data, the interaction behavior trajectory between customers and clothing items in the display area is extracted, and a spatial disturbance index representing the intensity of physical disturbance generated in the display area under customer interaction is calculated. Based on the interactive behavior trajectory, a dynamic interactive topology graph representing the spatiotemporal correlation strength between different clothing items is constructed, and a graph cohesion coefficient representing the degree of node association cohesion in the dynamic interactive topology graph is calculated. Calculate the time-varying slope of the spatial disturbance index within the current preset step size period. and the time-varying slope of the graph's cohesion coefficient. ; Determine the If the value is less than zero, then the current display state of the display area is determined to be in a disturbance attenuation state; otherwise: based on a preset nonlinear mapping function, the value is determined to be less than zero. With the Coupled calculations are performed to obtain the display conversion potential energy index; based on the numerical range of the display conversion potential energy index, the display state is determined to be either a structured interaction state or an unstructured interaction state. Generate and output corresponding display maintenance strategies based on the display status; Among them, Within the constraint range, the display conversion potential energy index and Positively correlated with It shows a negative correlation.

2. The method for evaluating and optimizing clothing display effects based on visual analysis according to claim 1, characterized in that, The calculation of the space disturbance index includes: The display area is divided into several sub-space units; Based on the depth point cloud data and the texture feature data, extract the height variance of the depth point cloud and the distribution complexity of the texture gradient within each subspace unit; The height variance and the distribution complexity are weighted and fused based on a preset mapping function to obtain a state value that characterizes the physical deformation features of a single subspace unit. The current distribution is obtained by statistically analyzing the probability distribution of the state values ​​of the plurality of subspace units within a preset value range. The relative entropy of the current distribution relative to the preset reference distribution is calculated as the spatial perturbation index.

3. The method for evaluating and optimizing clothing display effects based on visual analysis according to claim 1, characterized in that, The construction of the dynamic interactive topology graph includes: Based on the interactive behavior trajectory, the time sequence switching trajectory of the customer touching different clothing items in different areas within a preset time window is extracted. Construct an initial topology graph, where nodes represent clothing items touched by customers and edges represent the corresponding temporal switching trajectories; The intensity of the interactive action is determined based on the duration of the time-series switching trajectory in the area where the corresponding clothing item is located, and the rate of change of motion at the endpoints of the time-series switching trajectory. The weights of the corresponding edges in the initial topology graph are assigned in real time based on the intensity of the interaction action to obtain the dynamic interactive topology graph.

4. The method for evaluating and optimizing clothing display effects based on visual analysis according to claim 1, characterized in that, The formula for calculating the display conversion potential energy index is as follows: in, The preset intention gain coefficient, The preset disturbance attenuation penalty coefficient, It is a natural constant. The display transformation potential energy index.

5. The method for evaluating and optimizing clothing display effects based on visual analysis according to claim 1, characterized in that, The determination of whether the display state is a structured interaction state or an unstructured interaction state includes: Determine whether the display conversion potential energy index is greater than or equal to the preset high potential energy conversion threshold. If so, determine that the display state is the structured interaction state. Determine whether the display conversion potential energy index is less than a preset low potential energy conversion threshold, and whether the absolute value of the spatial disturbance index within the current preset step period is greater than a preset disturbance threshold. If so, determine that the display state is the unstructured interaction state.

6. The method for evaluating and optimizing clothing display effects based on visual analysis according to claim 2, characterized in that, The step of generating and outputting a corresponding display maintenance strategy based on the display status includes: If the display state is the structured interaction state, then a merchandising suppression strategy is output, which is used to indicate that merchandising in the current display area should be stopped. If the interaction is determined to be unstructured, an immediate maintenance strategy is output. The immediate maintenance strategy is used to instruct the clothing items with edge weights lower than a preset lower limit in the dynamic interaction topology graph to be sorted. If the disturbance decay state is determined, a merchandising suspension strategy is output. The merchandising suspension strategy is used to indicate that merchandising in the current display area should be stopped, and the preset reference distribution is updated according to the current visual time series data.

7. The method for evaluating and optimizing clothing display effects based on visual analysis according to claim 1, characterized in that, Also includes: Extract the target clothing item combinations in the dynamic interactive topology graph whose edge weights exceed a preset strong correlation threshold; Calculate the physical spatial distance between each garment item in the target garment combination within the current display area; If the physical space spacing exceeds the preset visual linkage distance, a layout reconstruction suggestion is generated. The layout reconstruction suggestion is used to indicate that the target clothing items are arranged in close proximity in the physical space.

8. A visual analysis-based apparel display effect evaluation and optimization system, applied to the display optimization of stacked apparel display areas, characterized in that... include: The data acquisition module is used to acquire visual temporal data of the display area, including continuous video frame streams, depth point cloud data and texture feature data. The data processing module is used to extract the interaction behavior trajectory between customers and clothing items in the display area based on the visual time series data, and to calculate the spatial disturbance index that characterizes the intensity of physical disturbance generated in the display area under the interaction of customers. The graph construction processing module is used to construct a dynamic interactive topology graph representing the spatiotemporal correlation strength between different clothing items based on the interactive behavior trajectory, and to calculate the graph cohesion coefficient representing the degree of node association cohesion of the dynamic interactive topology graph. The time-varying slope calculation module is used to calculate the time-varying slope of the spatial disturbance index within the current preset step size period. and the time-varying slope of the graph's cohesion coefficient. ; The status determination module is used to determine the status. If the value is less than zero, then the current display state of the display area is determined to be in a disturbance attenuation state; otherwise: based on a preset nonlinear mapping function, the value is determined to be less than zero. With the Coupled calculations are performed to obtain the display conversion potential energy index; based on the numerical range of the display conversion potential energy index, the display state is determined to be either a structured interaction state or an unstructured interaction state. The strategy generation module is used to generate and output corresponding display maintenance strategies based on the display status. Among them, Within the constraint range, the display conversion potential energy index and Positively correlated with It shows a negative correlation.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 7.

10. A computer storage medium storing computer-executable instructions thereon, characterized in that: When the computer-executable instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 7.

Images