Loading...
Please wait, while we are loading the content...
Similar Documents
Real-time Implementation of an Innovative Facial Database for Retail Customer Relationship Management ( CRM ) Systems
| Content Provider | Semantic Scholar |
|---|---|
| Author | Jiang, X. Li, Alex |
| Copyright Year | 2014 |
| Abstract | We built an automatic facial database in which a person's gender, facial characteristics and time of visit can be extracted and stored into the online database in real time. Such an application has the potential to increase in-store conversion rate in the retail sector. Based on this prototype system, further improvements could be made to develop a query system to push notifications to shop assistant's mobile devices, instructing them to make appropriate discounts on the right product, to the right customer, at the right time, based on the customers' visit and purchase history. The system is demonstrated to be a fast and storage efficient implementation, making use of a wide array of the popular methods in recognition, eigenface, fisherface and local binary pattern histogram. We compared, experimented, selected the best methods for the functions, and developed easy and effective mechanisms to achieve desirable system performance. Section 1: Introduction Recenlty, the offline retail sector has started to feel increasing pressure from online eCommerce. As the recent Forrester reports show, eCommerce will overtake the brickand-mortar store by 2014 in revenue [1]. There is a need for more innovative ways for the traditional retailers to offer more and attractive products and promotions based on customers' preferences. The bottleneck is the low in-store conversion rate, and more, the inability to collect individual store visit – conversion data offline. We therefore propose a computer vision application that would extract the customers' biological information like gender, and store the facial features in the database in a compact way to identify, in real-time, the customer's visit pattern. The long-term goal would be (in future research) to use the database and real-time reporting, of an individual's purchasing behaviors to promote individualized purchases. In our system, we would like to handle potentially tens and thousands of people's images to be stored as facial feature values and compared against in the image database. And therefore, the real-time operations would define the system to have fast retrieval. Pre-processing the images; The images were preprocessed to be 165*120 sizes and later reduced even further by Principal Component Analysis (PCA). Grayscale images were used to reduce 1 dimension from the original RGB images and also to make the images more invariant towards light; Tracking customers through reasonable estimate of speeds In our implementation, instead of using tracker algorithm to identify whether the person appeared in the video frame is the same one as before, we used reasonable estimation of shift in position, taking into account of the additional new faces into the view. With such a simple algorithm, the processing speed is fast, with tolerable level of error. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://cvgl.stanford.edu/teaching/cs231a_winter1415/prev/projects/crm.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
