Quantifying human behaviour in a retail environment.
Thesis DisciplineComputer Science
Degree GrantorUniversity of Canterbury
Degree NameMaster of Science
Robustly quantifying human behaviour in a retail environment raises research challenges around accurately and reliably recognising motion, age, gender, repeat customers and product acquisition in such unconstrained conditions. The motivation for this research is that computer vision can be used in the retail refrigeration industry to provide the shop/product owners with information about their clients, products, sales, stock levels and can also help with understanding the customers’ needs and psychology. This proposed method improves the accuracy of traditional face detection and recognition using depth information, in uncontrolled lighting environments and where the orientation of faces are not only front facing. Further proposed algorithms are tested on product recognition from a retail refrigeration unit in a retail setting. These proposed methods adapt Hue manifold, Haar cascade classifiers, SIFT and Local Binary Patterns Histograms. The face detection results of 96% recall and 100% precision together with face recognition results of 85% recall and 97%, indicates that the proposed method may be useful for improving face recognition in variable lighting environments where people do not stop moving and are not always facing the camera.