Context Driven Focus of Attention for Object Detection
Context plays an important role in general scene perception. In particular, it can provide cues about an object’s location within an image. In computer vision, object detectors typically ignore this information. We tackle this problem by presenting a conc
- PDF / 3,979,828 Bytes
- 18 Pages / 430 x 660 pts Page_size
- 93 Downloads / 236 Views
Abstract. Context plays an important role in general scene perception. In particular, it can provide cues about an object’s location within an image. In computer vision, object detectors typically ignore this information. We tackle this problem by presenting a concept of how to extract and learn contextual information from examples. This context is then used to calculate a focus of attention, that represents a prior for object detection. State-of-the-art local appearance-based object detection methods are then applied on selected parts of the image only. We demonstrate the performance of this approach on the task of pedestrian detection in urban scenes using a demanding image database. Results show that context awareness provides complementary information over pure local appearance-based processing. In addition, it cuts down the search complexity and increases the robustness of object detection.
1
Introduction
In the real world there exists a strong relationship between the environment and the objects that can be found within it. Experiments with scene perception, interpretation, and understanding have shown that the human visual system extensively uses these relationships to make object detection and recognition more reliable [1,2,3]. In the proper context, humans can identify a given object in a scene, even if they would not normally recognize the same object when it is presented in isolation. The limitation of a local appearance being too vague is resolved by using contextual information and by applying a reasoning mechanism to identify the object of interest. An example is shown in Fig. 1, where most people have little trouble in recognizing the marked objects in the image. However, shown in isolation, an indisputable recognition of these patches is not easily achieved. In general, context plays a useful role in object detection in at least two ways. First, it helps detection when local intrinsic information about the object is insufficient. Second, even when local appearance-based object detection is possible, the search space can be cut down by attending to image regions where the occurrence of the objects of interest is most likely. For example, when searching for manholes in Fig. 1 the search can be constrained to the ground plane. Even though object detection is a well established discipline in computer vision and is used in a large number of applications, contextual information is L. Paletta and E. Rome (Eds.): WAPCV 2007, LNAI 4840, pp. 216–233, 2007. c Springer-Verlag Berlin Heidelberg 2007
Context Driven Focus of Attention for Object Detection
217
Fig. 1. The object hypothesis formed from local appearance is rather weak for a unique object recognition. Using the surroundings of the patches significantly aids recognition.
typically ignored. Many concepts for object detection have been developed where, independent of the particular representation model used, the employed object detector is based on local appearance alone (see e.g. [4] for a review). Standard representation models are bag-of-feature
Data Loading...
