基于隐条件随机场的行人属性检测方法研究

Pedestrian attributes classification on surveillance environment

行人属性是指人的一些外部特征,如"发型","服饰","携带品"等。文中提出一种基于隐条件随机场的方法来发现行人属性。首先,对图像进行分割,其次对分割区域进行特征提取,然后利用隐条件随机场检测行人拥有的属性,并提供属性的具体位置。实验证明,"背包"、"短裤"、"牛仔裤"三种属性的检测率超过了50%。

The definition of pedestrian attributes is some appearance feature,like hairstyle,cloth and carry-things. This paper presents a hidden conditional random field approach to discover local attributes of pedestrian images. The approach first segments images into several regions,and then corresponding features are extracted from those regions. After that a hidden conditional random field model is applied to discover candidate local attributes. Finally,regions corresponding to local attributes are given. The experiments demonstrate that detection accuracy of some attributes like ‘backpack',‘shorts',‘jeans' is more than fifty percent.