Moving object detection device and object detection method

1. An object detection device comprising:

a candidate object selection unit suitable for converting a stereo image transmitted from moving cameras into a disparity image, and selecting pixels having a disparity value greater than a threshold value in histogram of the disparity image as candidate objects;

a global motion estimation unit suitable for calculating motion vectors for coordinates of the stereo image through optical flow, and estimating global motions representing movement of the moving cameras through sample data extracted from the motion vectors; and

an object determination unit suitable for extracting outliers from difference between global motion vectors calculated through the global motions estimated by the global motion estimation unit and the motion vectors calculated through the optical flow, and detecting a candidate object having a ratio of outliers greater than a predetermined ratio among the candidate objects selected by the candidate object selection unit as a moving object.

2. The object detection device of claim 1 , wherein the candidate object selection unit includes:

an object extraction unit suitable for converting the stereo image into the disparity image, and extracting pixels having the disparity value greater than the threshold value in histogram of the disparity image as objects; and

a distance value calculation unit suitable for calculating distance values of the extracted objects through the disparity image, and selecting the candidate objects among the extracted objects, the distance values of which are less than or equal to a predetermined value.

3. The object detection device of claim 2 , wherein the candidate object selection unit further includes a label unit suitable for marking a label on each of the candidate objects.

4. The object detection device of claim 1 , wherein the global motion estimation unit includes:

a motion vector calculation unit suitable for calculating motion vectors for coordinates of the stereo image through the optical flow; and

a motion vector determination unit suitable for calculating vector errors from the motion vectors, and estimating the global motions using the sample data extracted from the motion vectors, the vector errors of which fall within a preset range.

5. The object detection device of claim 4 , wherein the motion vector calculation unit calculates the motion vectors using Random Sample Consensus (RANSAC).

6. The object detection device of claim 1 , wherein the object determination unit includes:

an outlier extraction unit suitable for extracting outliers of the candidate objects from the difference between the global motion vectors and the motion vectors; and

an outlier reliability calculation unit suitable for detecting the candidate object having the ratio of outliers greater than the predetermined ratio among the candidate objects as the moving object.

7. An object detection method comprising:

converting a stereo image transmitted from moving cameras into a disparity image by a candidate object selection unit;

selecting pixels having a disparity value greater than a threshold value in histogram of the disparity image as candidate objects by the candidate object selection unit;

calculating motion vectors for coordinates of the stereo image through optical flow by a global motion estimation unit;

estimating global motions representing movement of the moving cameras through sample data extracted from the motion vectors by the global motion estimation unit;

extracting outliers from difference between global motion vectors calculated through the global motions and the motion vectors calculated through the optical flow by an object determination unit; and

detecting a candidate object having a ratio of outliers greater than a predetermined ratio among the candidate objects as a moving object by the object determination unit.

8. The object detection device of claim 7 , wherein the global motions are estimated through Random Sample Consensus (RANSAC).