以中心为主且适合有效率层级目标检测的评估外文文献翻译英文原文.docVIP

IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2004 Feature-Centric Evaluation for Efficient Cascaded Object Detection Henry Schneiderman 1 Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213, USA hws@ Abstract where the classifier evaluates all possible “windows” in the image as shown by the rectangles. The drawback of this approach is that such an exhaustive search can be very time consuming. We describe a cascaded method for object detection. This approach uses a novel organization of the first cascade stage called “feature-centric” evaluation which re-uses feature evaluations across multiple candidate Search in scale windows. We minimize the cost of this evaluation through several simplifications: (1) localized lighting normalization, (2) representation of the classifier as an additive model and (3) discrete-valued features. Such a method also incorporates a unique feature Search in location representation. The early stages in the cascade use simple fast feature evaluations and the later stages use more complex discriminative features. In particular, we propose features based on sparse coding and ordinal relationships among filter responses. This combination of cascaded feature-centric evaluation with features of Figure 1. Exhaustive search object detection One way of reducing the computational cost of exhaustive search is to express the classifier as a cascade of sub-classifiers as shown in Figure 2. Each sub- classifier stage makes a decision to either reject the input, classifying it as non-object or to continue evaluation using the next sub-classifier stage. Those windows that survive the last sub-classifier are classified as the object. This strategy is designed to remove the vast majority of increasing complexity achieves both computational efficiency an