We develop an object detection method combining top-down recognition with bottom-up image segmentation. There are two main steps in this method: a hypothesis generation step and a verification step. In the top-down hypothesis generation step, we design an improved Shape Context feature, which is more robust to object deformation and background clutter. The improved Shape Context is used to generate a set of hypotheses of object locations and figureground masks, which have high recall and low precision rate. In the verification step, we first compute a set of feasible segmentations that are consistent with top-down object hypotheses, then we propose a False Positive Pruning(FPP) procedure to prune out false positives. We exploit the fact that false positive regions typically do not align with any feasible image segmentation. Experiments show that this simple framework is capable of achieving both high recall and high precision with only a few positive training examples and that this method can be generalized to many object classes.
|
|
 |
[1] Object Detection Combining Recognition and Segmentation, Liming Wang, Jianbo Shi, Gang Song, I-Fan Shen, Eighth Asian Conference on Computer Vision (ACCV), 2007 [pdf][Poster] [Slides (the first part)] |
 |
Penn-Fudan Pedstrian Detection and Segmentation This is an image database containing images that are used for pedestrian detection in the experiments reported in [1]. The images are taken from scenes around campus and urban street. The objects we are interested in these images are pedestrians. Each image will have at least one pedestrian in it. The heights of labeled pedestrians in this database fall into [180,390] pixels. All labeled pedestrians are straight up. There are 170 images with 345 labeled pedestrians, among which 96 images are taken from around University of Pennsylvania, and other 74 are taken from around Fudan University. |
 |
This code is written for the paper [1]. It was initially developed by Gang and Liming, and cleaned up by Liming in Jan 2008. It's tested under Matlab 7.0. The version number is now 1.0beta. The main part of the code uses improved Shape Context as feature descriptor and fit into a Hough Voting framework to detect objects. It can be used for initial hypothesis proposal. I hope you find this code helpful! Go to source code page or download it. |
 |
Go to more results page. |
 |
What does Wukong see? |