WO2009072000A2

WO2009072000A2 - Fit prediction methods for virtual fitting of footwear to a customer

Info

Publication number: WO2009072000A2
Application number: PCT/IB2008/003812
Authority: WO
Inventors: Wei Shi
Original assignee: Wei Shi
Priority date: 2007-09-14
Filing date: 2008-09-12
Publication date: 2009-06-11
Also published as: WO2009072000A3

Abstract

Disclosed is a method to predict the fit of a show on a user, a virtual fitting of footwear to a customer. The present invention provides an accurate prediction of a fit of a certain piece of footwear to a certain foot without an actual tryout. This process may be regarded as a virtual fitting since a customer will not need to physically try on a pair of shoes to determine whether not they are a good fit. When a customer orders from a printed catalog, TV shopping channel, an online shopping website, or in other situation when it is not possible to try on shoes, it is important to predict fit quality of footwear to foot.

Description

FIT PREDICTION METHODS FOR VIRTUAL FITTING OF FOOTWEAR TO A

CUSTOMER

This application claims priority to U.S. Provisional Application 60972622 filed 14- SEP-2007, the entire disclosure of which is incorporated by reference.

TECHNICAL FIELD AND BACKGROUND

The present invention relates to Fitting of Footwear to a Customer. More specifically, the present invention provides a way to accurately predict fit quality of a certain piece of footwear to a certain foot without an actual tryout. The present invention is to accurately predict fit quality of a certain piece of footwear to a certain foot without an actual tryout. This process may be regarded as a virtual fitting since a customer will not need to physically try on a pair of shoes to determine whether not they are a good fit. When a customer orders from a printed catalog, TV shopping channel, an online shopping website, or in other situation when it is not possible to try on shoes, it is important to predict fit quality of footwear to foot. Alternatively, in a brick-and-motor footwear retail location, for a customer to try on all shoes in his/her shoe size and favorable style may be time consuming; Fit prediction becomes a useful pre-screening mechanism to exclude shoes that are very likely to be a bad fit, and recommend shoes likely to be good fit, hence improve sales efficiency. The present invention is a set of new set of methods to predict fit quality of footwear. A database is constructed and populated by customer records. Besides personal and contact information, foot shape geometry and related measurements, each customer's record will contain physical try-on fitting history of this customer to various shoes. An algorithm is used to search the database to locate customers with the most similar physical try-on fitting history.

Each customer's record will contain a similarity list of other customers in the order of similarity in fitting history. When a customer has interest in a certain style of shoe, a query is submitted to the database. Through fit quality prediction methods proposed in this invention, fit quality can be predicted with a high degree of accuracy and consistency. The central ideas of the invention may be summed up as:

1. having other people with similar past fitting history to physically try on the shoes you are interested in for you, and relate their fitting experience to you, and

2. based on a small number of physical try-on comfort rating, generate a representative foot geometry for a shoe. Comparing this representative foot to the foot of prospective customers to predict fit quality. 3. based on a small number of physical try-on comfort rating, generate a representative shoe last geometry for a foot in a given style of shoe, such as dress, sports, high-heel. Comparing this representative shoe last to the shoe last of prospective shoe in the same style to predict fit quality. When the size of the database grows above a critical threshold (in terms of customer physical try-on comfort ratings), most people will no longer need to try on shoes. After a new style of shoes goes on sale, each size in this style will need to be physically tried on by only a small number of customers. Physical try on fit quality of these small number of customers will suffice to accurately predict fit quality of this style of shoes to other customers through the similarity list built in each customer's record in the database.

How your invention may be used differently:

The underline principal of the invention can be applied readily to: garment and eyewear fit prediction and any other instances where an item is fit to a human and the fit quality is important but difficult to predict without an actual physical try-on.

DETAILED DESCRIPTION

Various aspects of the embodiments will be described using terms commonly employed by those skilled in the art to convey the substance of their work to others skilled in the art.

Various operations will be described as multiple discrete operations, in turn, in a manner that is most helpful in understanding the present invention, however, the order of description should not be construed as to imply that these operations are necessarily order dependent. In particular, these operations need not be performed in the order of presentation.

1. Obtaining Customer Foot Measurement: The number of foot measurements pertains to footwear fit prediction is anywhere between 1 and 50 or more. In the simplest example, foot length is used. In a more sophisticated model, foot length, width, toe height, medial malleolus height, ankle circumference, ball girth, and other linear or circumferential measurements are also included. Foot measurements can be obtained by various means, such as a tape measure, Brannock device, or any other measurement means. Recent years has seen a proliferation of non-contact 3D foot digitizing tools, which are more efficient than traditional ways. From 3D digital model, one can obtain all desired foot measurements. A 3D foot is a 3D digital model of a customer's foot acquired by various 3D foot scanning machines commercially available, such as the Yeti 3D foot scanner made by Vorum Research Corporation of Canada.

Customer 3D foot can be stored in various 3D file formats. Alternatively, only foot measurements can be stored, which will save storage space. Later on when a customer 3D foot is required, it can be reconstructed from the stored foot measurements through a technique known as "morphing" from a standard 3D foot model 2. Obtaining Shoe Last Geometry:

Shoe last is the solid mould around which shoes are fabricated, including mass- produced shoes and bespoke shoes. Shoe lasts are usually made of plastics, wood, or metal. Shoe lasts provide the only reliable and non-deformable inner-space information about the shape of the shoe. One can say that the shoe last is the soul of the shoe and ultimately determines whether a shoe will fit a foot. With the wide-spread use of CAD systems, nowadays many shoe lasts are designed in a CAD environment and hence the 3D digital model of shoe last is quite often available, or alternatively the 3D digital model of shoe lasts can be acquired by various 3D scanning solutions. Therefore shoe last measurements can be extracted from 3D model or measured from the physical shoe last itself.

Shoe last 3D model can be stored in various 3D file formats. Alternatively, only shoe last measurements can be stored, which will save storage space. Later on when a shoe last 3D model is required, it can be reconstructed from the stored shoe last measurements through a technique known as "morphing" from a standard shoe last model.

3. Fit Quality of a Foot to a Shoe:

Fit quality is a subjective matter and can be represented in many ways. Fuzzy theory provides means to calculate and compare subjective matters such as fit quality.

For illustration purpose, it is helpful to assigning a comfort rating number. Fit quality may be designated as Excellent Fit=1 , Good Fit=0.85, Acceptable Fit=0.7, Bad Fit=0.3, and Not Fit At AII=O. A customer may also assign a comfort rating number anywhere in between these numbers. For example, a customer may feel that a certain shoe has a comfort rating better than "Good Fit" but not quite as good as "Excellent Fit", hence assigning a comfort rating of 0.9.

More elaborate fit quality representation will yield more accurate fit prediction results but will require more input from customer. For example, one can imagine fit quality can be defined as tightness/looseness comfort at various part of the foot during different stages (standing, walking, running, etc.).

Comfort rating is the only subjective judgment step in this invention and is subject to inconsistency between customers and even within the fitting experience of one customer over time. The subjective dimension of shoe fitting may be eliminated when there exist a way to measure pressure over the entire surface of the foot dynamically over the entire gait cycle during walking, running, standing, sitting, etc.. This goal may be achievable as there are companies supplying dynamic gait pressure products, only they are too bulky for in-shoe use. I envision a thin "pressure sock" that can dynamically capture pressure information and transmit by wire or wirelessly to a computer for analysis. With this device, we can establish an objective measure of comfort for shoe fitting.

4. Comparing Foot to Shoe Last to Predict Fit Quality:

If customer foot measurement and shoe last geometry are both stored in the database and available for comparison, an objective fit quality can be generated by comparing the two. In one example, foot measurements and corresponding shoe last measurements are compared to derive a fit quality. In another example, 3D foot model and 3D shoe last are overlaid upon each other and spatial differences are calculated to derive a fit quality. Comparing a 3D foot model to a 3D shoe last is the first step to accurately predict fit quality.

This is a major step forward comparing to merely using two numbers (length and width) to determine shoe fit. With 3D foot model and 3D shoe last model, scores of measurements can be compared to determine an objective fit quality. There are many prior art literature describing steps involved in comparing 3D foot shape to a 3D shoe last model, and deriving a custom shoe last from foot shape. Basically, a 3D shoe last will be tighter at certain locations by a certain amount and looser at other locations by a certain amount.

However, the fitting quality of a shoe to a foot is ultimately a subjective matter. Different customers will have different preference for tightness/looseness at different part of the foot.

Therefore two customers having almost identical 3D foot model may have different preference for shoe size. By the same token, two customers with different 3D foot geometry may have found the same shoe to be good fit.

To complicate the issue further, most of the 3D foot scan solutions available today only take a 3D foot model when a customer is standing statically on a flat surface. But the foot is part of a living human anatomy, it deforms during walking, standing, running to different degrees.

There may be commercially available 3D scanning solutions in the near future that can take a 3D foot model during different status such as walking, running.

A shoe last is generated based on 3D foot plus heel-height, toe-spring and toe box shape adjustments. Different style of shoes will have different heel-height, toe-spring and toe-box shape, resulting in different shoe last geometry for the same customer.

Therefore, fit prediction by comparing customer foot to a shoe last of the desired style of shoe is not enough.

5. Comparing Fit history Between Customers to Generate Similarity Ranking List for Each Customer.

Each customer's record will contain physical try-on fitting history of this customer to various shoes. An example would be an entry of Nike Air, Model XXX, Size 10-E, Fit quality is good, comfort rating 0.85.

Search the database to locate customers with the most similar physical try-on fitting history.

An example: For Customer A, throughout the entire database, customer B shares the most similar fitting history. Each customer's record will contain a similarity ranking list of other customers in the order of similarity in fitting history. Similarity in fitting history may be determined by calculating a similarity score as follows. If customer A and B have tried on the same brand, style and size of shoe and found them to be of same fit quality, similarity score is increased by 1 point. If customer A and B have tried on the same brand, style and size of shoe and found them to be of opposite fit quality (i.e. customer A find the shoe to be excellent fit with a comfort rating of 1 , but customer B find the same shoe to be not fit at all with a comfort rating of 0), similarity score is decreased by 1 point. If there is 0.5 point difference separating customer A and B's comfort rating to a certain shoe, similarity score remain unchanged. Generally, if there is X point (0DXD 1 ) difference separating customer A and B's comfort rating to a certain shoe, similarity score is changed by 1 -2^*X (one minus two times X). A positive change is increase while a negative change is decrease. Each customer's fit history will be compared to all other customers. For example, customer A's fit history is compared to all other customers, but only those customers who have physically tried on and recorded comfort rating of at least one shoe that is the same as customer A will proceed to generate a similarity score. The similarity score is used to rank all these customers into customer A's similarity list. Customer A's similarity ranking list will contain all other customers who have a similarity score of at least 1 in relation to customer A. The database and each customer's record are constantly being updated as new customers are being added and new physical fitting are performed and comfort rating are recorded.

6. Use a Customer's Similarity Ranking List to Predict Fit Quality

When a customer has interest in a certain style of shoe, a query is submitted to the database. A search is performed in the similarity list of the customer's record to find other customers who have already did a physical try-on for the shoe in question. These "other customers" comfort rating to the shoe can be used to predict fit quality of the customer in question as follows.

Suppose customer A wants to determine fit quality of shoe B to his foot. A query is submitted to the database and found 3 other customers (Let's call them customer C, D, and E) in customer A's similarity list have physically tried and recorded comfort rating for shoe B. Customer C, D, and E each have a similarity score of 5, 3, and 1 respectively in customer A's similarity ranking list. And customer C, D, and E each have a comfort rating of 0.85, 1 , and 0.3 for shoe B respectively. The comfort rating of customer A to shoe B may be estimated as a weighted-average of comfort ratings of customer C, D, and E to shoe B, with similarity score as weight. In this example, customer A's comfort rating for shoe B may be predicted as (0.85X5+1 X3+0.3X1 )/(5+3+1 )=0.833, or slightly below "good fit".

For this approach to work, there must be at least one customer in customer A's similarity ranking list who have tried on shoe B and recorded comfort rating. If there is only one customer in customer A's similarity ranking list (similarity score of 4) who have tried on shoe B and recorded comfort rating of 0.7 customer A's comfort rating for shoe B can be predicted as (0.7X4)/4=0.7, or an "acceptable fit".

This approach is particularly useful when neither customer foot measurements nor shoe last geometry is available. Even when customer foot measurements and shoe last geometry are available, this approach will generate a more reliable fit quality prediction. It may be regarded as having other people with similar past fitting history to physically try on the shoes you are interested in for you. The prediction is more subjective and hence more reliable. 7. Generating Representative Foot Geometry for a Particular Shoe.

When a shoe has been physically tried on (with comfort rating recorded) by at least one customer whose foot measurements are recorded, representative foot geometry can be generated by a weighted-average method. For example, after 5 customers have physically tried on the shoe, each will have a comfort rating for this shoe. Presume that these customers' foot measurements are available; the representative foot geometry can be derived by weighted-averaging of corresponding measurement. For instance, the ball girth of the representative foot would be the weighted average of the ball girth measurement of these 5 customers, with the comfort rating as the weight. The representative foot for a particular shoe is constantly adjusted as new customers try on the shoe and more customers are added into the calculation for representative foot. 8. Comparing a Representative Foot of a Shoe to a Customer foot to Predict Fit Quality. Due to the difficulties associated with comparing directly foot measurements with shoe last geometry, as mentioned above, it is desirable to limit this practice to the beginning of a style of shoe's marketing cycle, when no customer have tried on the shoe and hence no comfort rating is available. Once a few customers have tried on the shoe then a representative foot can be generated for this shoe. Later customers' foot will no longer be compared with the shoe last, but with the representative foot to predict fit quality. In one example, representative foot measurements and customer foot measurements are compared to derive a fit quality. In another example, 3D representative foot model and 3D customer foot model are overlaid upon each other and spatial differences are calculated to derive a fit quality. This method has the benefit of not requiring shoe last geometry information. Only foot geometry is required.

9. Generating Representative Shoe Last Geometry for a Particular Foot for a Particular Style of Shoe An example: customer A's fitting history contains two pair of dress shoes C and D. Now customer A needs to do a virtual fitting of a dress shoe E. Shoes C, D, and E all have their shoe last geometry available. Customer A's comfort rating for C and D are 0.85 and 1 respectively. Shoe lasts for C and D can be combined into a representative shoe last for dress shoe for customer A. One embodiment is the weighted-average method to calculate measurements of the representative shoe last from C and D with comfort rating as weights.

The representative shoe last for a customer in a given style of shoe is constantly adjusted as the customers try on the shoes and record fit quality comfort rating. As sated above, a custom shoe last is derived from foot measurements, toe-box design, and heel height design. Therefore one must be careful when deriving a representative shoe last from shoe lasts of past tried-on shoes. In any particular style (such as dress, sports, boots, etc.), only shoe lasts of similar heel heights and toe-box shape should be used to generate representative shoe last. This does not detract the value of this method, since there are a very limited number of heel heights and toe-box designs. Example: Customer A has tried on three pair of dress shoes of pointed toe and heel height of 1 inch C, D, and E with comfort ratings recorded. A representative shoe last is generated from shoe lasts of shoe C, D, and E through weighted-average method. The representative shoe last becomes the customer's representative shoe last in the pointed toe design with a heel height of 1 inch. Later on, when customer A wishes to virtual fit another pair of dress shoe of pointed toe design and heel height of 1 inch, the representative shoe last can be compared to the shoe last of the target shoe to predict fit quality. The representative shoe last may be used to fabricate bespoke custom-made shoe for the customer, but at least one of the shoe lasts used to generate the representative shoe last should have an "Excellent Fit" rating with a comfort rating of 1. This method is superior to generating shoe last based on foot measurements because the representative shoe last takes into account fit comfort ratings of previous try-ons in a particular style. However, due to the complex nature of fit quality of shoes, the shoe manufacturer may choose to combine the foot geometry information of customer (if available) and the representative shoe last geometry to derive a bespoke shoe last. The representative shoe last is a virtual shoe last that does not exist in solid form until a customer requires fully custom-made shoes. For custom selection of footwear, a virtual shoe last is enough to facilitate fit quality prediction. 10. Comparing Representative Shoe Last Geometry to the Last of a Target Shoe to Predict Fit Quality.

After a customer record contains a number of fitting quality comfort ratings, and the shoe last geometry of past fitting and the desired shoe are available, a comparison of the shoe last of the desired shoe and the shoe last of a same style of shoe from past fitting history can predict fit quality.

In one example, representative shoe last measurements and desired shoe last measurements are compared to derive a fit quality. In another example, 3D representative shoe last model and 3D desired shoe last model are overlaid upon each other and spatial differences are calculated to derive a fit quality.

This method has the benefit of not requiring a customer's foot measurement information.

Only shoe last geometry is required. How do these components or steps work together, and how is the invention used:

1. As described in the previous section:

Step 1 is used acquire customer foot measurements.

Step 2 is used to acquire shoe last geometry.

Step 3 defines fit quality of a foot to a shoe Step 4 is used to predict fit quality by direct comparison of a foot with a shoe.

Step 5 generates similarity ranking list for each customer

Step 6 is used to predict fit quality by similarity ranking list of a customer

Step 7 generates representative foot geometry for a shoe

Step 8 is used to predict fit quality by comparison of representative foot geometry of a shoe to a customer foot.

Step 9 generates representative shoe last for a foot in a particular style of shoe.

Step 10 is used to predict fit quality by comparison of representative shoe last to the shoe last of a desired shoe. Steps 1 , 2, 3, 5, 7, 9 are preparatory steps, and steps 4, 6, 8, 10 are actual fit prediction methods. Steps 5, 6, 7, 8, 9, 10 and their combination seem to be novel and patentable. Step 4 is used when both customer foot measurements and shoe last geometry are available. This step yields a rough prediction of fit quality for reasons described above. This step can be used as a pre-screening to exclude obvious bad fit between a foot and a shoe but cannot be counted to reliably predict fit quality. This step is useful at the beginning stage of the database when there is not enough customer record and similarity ranking list, so that step 6, 8, and 10 cannot be applied. But Step 4 requires both customer foot measurement and target shoe last geometry to do the comparison.

Step 6 is used to predict fit quality when customer record and similarity ranking list reaches a critical mass. This step can be used without customer foot measurements and shoe last geometry information. This step yield a highly reliable fit quality prediction due to the fact that it is based on the fitting quality comfort rating of other customers who share similar fitting history and who have tried the shoe. This method yields a subjective fit quality prediction and takes into account all subjective fit factors that are hard to define and hard to measure.

Step 8 is used to predict fit quality when a shoe has been physically tried on (with comfort rating recorded) by at least one customer whose foot measurements are recorded. This step can be used without shoe last geometry but must have customers' foot measurements to the desired shoe. Step 10 is used to predict fit quality when a customer has been physically tried on (with comfort rating recorded) on at least one style of shoe and wish to do a virtual fit another shoe in the same style. This step can be used without customer foot measurements but must have shoe last geometry of past tried-on shoes and the present desired shoe in the desired style. Step 4, 6, 8, and 10 may be collectively applied to arrive at a more reliable fit prediction.

2. When a new style of shoe goes on sale, it is not a finished product. Because the new style of shoe in different size have not been tried out by any customers. Even when the shoe last geometry is available and a large number of customers have foot measurements stored in the database, only a rough prediction of fit quality can be obtained for these customers, because the inherent limitations of predicting fit quality by way of comparing foot measurements with shoe last geometry as stated above. To accurately predict fit quality, a subjective dimension must be introduced, and that is the aim of the present invention. Steps 6, 8, and 10 as described above bring in a subjective criterion by horizontally compare comfort rating of different customers and vertically compare comfort rating of different shoes of the same customer.

3. Mass-customization has been a hot field of research in recent decades. It has been a dream waiting to be fulfilled for the whole footwear industry. However mass- customization of shoes has not made significant progress in recent years. The problem is that currently there exits no reliable way to predict the fitting quality of a foot to a shoe. Hence a customized shoe last cannot be generated automatically by software to facilitate mass-customization of shoes.

4. Another problem hindering the development of mass-customization of shoes is lack of customer foot measurement data (including 3D foot models) and shoe last geometry information. The cause of this problem is that for the vast majority of customers there exist no incentive for them to have their foot measurements recorded, if they are not immediately interested in custom-made shoes. As stated in section 2 above, using the present invention, a customer can realize immediate benefit in online-purchasing (improved fit confidence) or in store-buying (No need to try all shoes in your size, many will be excluded by methods proposed). As the size of the database grow, the shoe industry will eventually retool its process to push forward mass-customization of shoes.

5. Shoe shopping in retail locations as a leisure activity will never go away; part of the population will always enjoy the experience of going into a store and physically try on many shoes before finding an excellent fit. These fitting histories should be recorded not only to the benefit of this customer but also other customers.

6. Incentive for trying on shoes: As sated above that a new shoe in a particular size going on sale is not a completely finished product; the manufacturer may offer incentive for customer to try on the new shoe and record comfort rating. Inventive steps 6, 8, and 10 of the proposed invention all require fit comfort rating for certain number of customers in the beginning of the marketing cycle.

The incentive will be high in the beginning and decrease as more people record their comfort rating. Because the contribution of later shoe try-on will not be as great as earlier try-on. The incentive will approach zero when adding more comfort rating of customers will not significantly improve fit quality prediction (as determined by feedback from customers who acquired shoes based on fit prediction methods). The rationale behind this incentive is that the customers first try on the shoes are helping them manufacturer complete a product (making the product easier to select and purchased with more confidence by other customers).

7. Incentive for providing feedback after physically try-on shoes suggested by fit quality prediction methods proposed. This feedback fit comfort rating should be recorded and used to improve fit quality prediction model.

8. The database can also be used to help shoe manufacturers and industry standard setting bodies analyze data and create more sensible sizing scheme. For example, why are shoe size spaced every inch? By data-mining foot measurements records, better shoe sizing scheming can be devised which means more people will find better fitting shoes, and manufacturers will have less unsold unwanted shoes because of improper fit.

9. The database can be used to extract customer preference in style, color and other fashion aspects as basis for developing and targeted marketing future products. Through data-mining, manufacturers can study trends in the shoe fashion industry.

Claims

CLAIMSWhat is claimed is:

1. A method to predict shoe fit comprising: recording fitting experiences of a shoe wearer and shoe to provide geometry information connected to the shoe wearer and the shoe; and matching a shoe buyer's geometry information with the shoe wearer's geometry information to produce a proper fit and size of the shoe for the shoe buyer to purchase without trying on the shoe.

2. The method of claim 1 wherein the geometry information connected to the shoe wearer and the shoe provide a comfort rating.

3. The method of claim 1 wherein the geometry information connected to the shoe wearer and the shoe is used to generate a representative foot shape of the shoe a prospective customer can use this representative foot shape to predict how the shoe will fit them.

4. The method of claim 1 wherein encouraging customer to provide feedback in the form of comfort rating, by offering certain incentives, after buying shoes according to fit prediction results, and apply these feedbacks to confirm the effectiveness of the fit prediction model or improve fit prediction model.

5. The method of claim 1 wherein encouraging physical try-on and comfort rate recording, at the beginning stage of a new shoe's marketing cycle, by offering certain types of incentives, in which the incentives decrease as more try-on are performed and more comfort ratings recorded, because physical try-on at the beginning of the marketing cycle of a new shoe style contribute more to the fit prediction model than later physical try-on.