| 摘要 |
An apparatus for detecting multiple objects by using adaptive block partitioning is disclosed. An object contour extraction unit extracts object contour information by using a local binary pattern (LBP) and a frame difference between subsequent images. An adaptive block partitioning unit adaptively partitions the object, which is not overlapped with other objects, into blocks on the basis of the extracted object contour information. A motion quantization unit calculates a motion orientation histogram (MOH) of the object by performing N-directional quantization for a motion vector. An object detection unit detects the object by using the blocks partitioned into an adaptive size, the object contour information, and the MOH, and estimates the labeling and movement direction of the detected object. According to the present invention, data can be effectively processed through eight-directional quantization of the object motion vector by using motion information given in advance by an ISP chip, a region suitable for the object of a block unit can be detected while minimizing an object motion error by using the block of the adaptive size and the orientation histogram, and the movement direction of the object can be estimated simultaneously with the detection of the object. |
