當推出產品或是展覽時，店家通常關心的是商品或是廣告對消費者的吸引程度，傳統上必須花費大量的資源與人力來調查消費行為，其中常見的是問卷或是消費紀錄來做分析。因此本計劃將提出一個以電腦視覺為基礎的消費者行為調查，將裝設的監控攝影機在需要分析的展覽櫥窗或是商品後方，直接取得消費者的影像，透過影像分析出性別與年紀，針對消費行為做出一些分析，像是何種年齡層的消費者喜歡目前櫥窗的商品。本計劃將系統分成三年來完成，第一年將包含兩個部分：商品櫥窗前人數統計及正臉偵測。主要是針對走道上的行人追蹤及計數，並且找出對應的正臉影像。第二年初步利用走路姿態的性別偵測與年紀估計，再結合前一年正臉偵測的部份達到更精確的消費者年齡和性別，目的是商品所在之人潮族群之分析，所以我們將直接以經過的人(包含經過或是有興趣之消費者)其走路姿態來做性別和年齡的偵測以供之後做為分析消費者感興趣與否之消費行為分析。在第三年的計畫中，我們預計採用三大類特徵來表示關注程度的高低；第一類是走路的行為，第二類是正臉面對櫥窗的時間，第三類是臉部的表情。根據這幾種不同的特徵組合，我們用以評量行人對於櫥窗內物品的關注程度。 Under the circumstance of high economic development society, tremendous resources and human effort have to be spent to investigate one’s buying behavior in selling a product or preparing a show. The reasons why need such resources and efforts are twofolds. One is the inefficiency of investigation mechanism which is usually accomplished by analyzing questionares or custom’s records. Another is the lacking of matching hardwares which is usually done by manual statistic analysis and data collection. In response to these needs, a vision-based custom purchasing behavior investigation system is proposed. In our system, a camera is mounted directly above the advertising merchandises to capture custom’s video sequence passing by. Through image analysis and pattern recognition, we can recognize the gender and age of consumers on which significant statistic data, such as the number of people staring at the merchandises, can be obtained.We plan to develop the system in three years. In the first year, we plan to accomplish two tasks including people counting and frontal face detection such that we can count the number of passengers who has interest to the merchandises. In the second year, we plan to analyze the walking behavior to recognize the gender and age of the passengers. In the meantime, the frontal faces of passengers who have interest in the merchandises will be captured. We will also analyze the frontal faces of passengers to confirm their gender and age. In the third year, three types of features will be analyzed to estimate the interest degree of the passengers who has interest to the merchandises. These three types of features include walking behavior, the period of eye staring, and facial expression. The whole system will be integrated in the third year also. 研究期間：10008 ~ 10107
