System and method for detecting still objects in images
The present invention provides an improved system and method for object detection with histogram of oriented gradient (HOG) based support vector machine (SVM). Specifically, the system provides a computational framework to stably detect still or not moving objects over a wide range of viewpoints. The framework includes providing a sensor input of images which are received by the “focus of attention” mechanism to identify the regions in the image that potentially contain the target objects. These regions are further computed to generate hypothesized objects, specifically generating selected regions containing the target object hypothesis with respect to their positions. Thereafter, these selected regions are verified by an extended HOG-based SVM classifier to generate the detected objects.
1. A method for detecting at least one still target object in at least one image comprising:
receiving a dataset of training samples in a computer, said training samples including images of still target object;
selecting regions in the training samples potentially having the still target object;
computing an extended histogram of oriented gradient (HOG) feature in an image patch of the selected region to generate an angle dimension and a distance dimension of each pixel in said image patch; and
training at least one classifier based on the computer extended HOG feature to detect the still target object in said image patch.
2. The method of claim 1 wherein said image patch is a rectangular window of pixels.
3. The method of claim 1 further comprising the step of generating position hypothesis of said selected regions, wherein said position hypothesis comprise potential still target object.
4. The method claim 2 wherein said generating step comprises obtaining depth map of the selected regions having the potential still target and calculating a height and density of each pixel of said depth map.
5. The method of claim 3 wherein said distance dimension is measured relative to a center pixel of said image patch.
6. The method of claim 1 wherein said extended HOG feature comprise a 2D orientation histogram displaying a position of the potential target object with respect to the image patch of the selected region.
7. The method of claim 1 wherein each of said classifier is trained according to a specific pose of said potential target objects.
8. The method of claim 1 wherein said classifier is a support vector machine (SVM) classifier.
9. The method of claim 8 wherein said SVM classifier is trained as
y
=
sign
(
∑
i
=
1
N
y
i
α
i
K
(
x
,
x
i
)
+
b
)
where x is the extended HOG feature of an observation example, yε{+1, −1} is a class label, x i is the extended HOG feature of i th training sample, N is the number of the training samples, and k(x,x i ) is the kernel function.
10. The method of claim 1 wherein said images are rendered images.
11. The method of claim 1 wherein said images are non-rendered images.
12. A computer program product comprising a non-transitory computer readable storage medium having a computer program stored thereon for performing the method of claim 1 .