| Summary: | 碩士 === 國立高雄第一科技大學 === 資訊管理系碩士班 === 105 === In view of the implementation of the Executive Yuan Business 4.0 strategy, we are facing the advent of a new era of business, in which the application of smart retailing is more important here. And hope to integrate the multi-channel path to Omni-Channel, combined with data analysis to establish consumer behavior model in order to enhance service value.
In this study, we collected the public open face database that containing gender and age label. We used Caffe Framework and GoogLeNet Network to training the gender and age models. In addition, we combined the skeleton tracking to locate face position and motion recognition via Kinect sensor. Finally, we used the historical order data to run association rule between the products that support by Nargo e-commerce. Then implement the product recommendation based on a customized customer groups.
In the experiment, we tested and compared the gender and age model with public open model (such as LAP model). The results show that in the public open face database (such as imdb-wiki 500k+ etc.) reach 86.12% and 76.39% accuracy rate. The average execution time is 1.04 seconds. Although it is not better than the LAP model, but it is more suitable for real-time application of this system. In part of the motion recognition, we used LV3D frame level feature and DTW method. The results show that in the cross-tested and unknown start and end motion can reach up to 90% accuracy rate. Finally, in the historical order data association rule analysis, we used the support values and confidence values to implementation of product recommendation. And we hope to bring more help for the enterprise in the marketing.
|
|---|
