Quantifying Contextual Information for Object Detection

Context is critical for minimising ambiguity in object de- tection. In this work, a novel context modelling framework is proposed without the need of any prior scene segmen- tation or context annotation. This is achieved by explor- ing a new polar geometric histogram descriptor for con- text representation. In order to quantify context, we for- mulate a new context risk function and a maximum margin context (MMC) model to solve the minimization problem of the risk function. Crucially, the usefulness and goodness of contextual information is evaluated directly and explic- itly through a discriminant context inference method and a context confidence function, so that only reliable con- textual information that is relevant to object detection is utilised. Experiments on PASCAL VOC2005 and i-LIDS datasets demonstrate that the proposed context modelling approach improves object detection significantly and out- performs a state-of-the-art alternative context model.

• Type: Conference
• Year: 2009
• Where: ICCV
• Authors: Wei-Shi Zheng, Shaogang Gong and Tao Xiang
• Keywords: Computer Vision, ICCV 2009, Object Detection