基于Hough树林的空间有形目标特征训练与检测识别方法

Space shape object feature training and detection and recognition method based on Hough forest

随着空间飞行器利用率的不断提高,各国开始关注于空间目标的监视问题。如何对空间有形目标进行准确的分类与定位识别是目前关注的难点之一。针对空间目标的特征提取与识别定位问题展开研究,提出了一种基于Hough树林的空间目标探测识别方法。首先,通过广义的Hough变换,使用独立的目标局部的探测识别对全局目标可能位置中心进行投票。然后,用与Hough图像极大值相对应的探测识别假设对局部特征的投票进行汇总,继而通过训练建立起Hough树林。进一步地,在传统码表存储局部特征投票信息的理论基础上,使用所建立的Hough树林进行了天基目标的探测识别。实验表明,此方法可以在不同的探测距离上对多类空间目标进行较好的探测识别。

As constant increasing utilization of space-aircraft, more and more countries begin to focus on space object surveillance. But how to precisely classify and localize for space shape object is one of difficulties. The problems of space object feature extraction, recognition and localization were studied and a space object detection and recognition method based on Hough forest was proposed. The method was accomplished via generalized Hough transform, in which voting for the possible locations of centroid of whole space objects with the detection of object parts; then these part of feature votes were accumulated with the recognition hypothesis which corresponded to the Hough image, and Hough forest was built by training stage. Furthermore, based on application of traditional codebook theory in saving feature-voting information, the space-based object detection and recognition were realized by using the established Hough forest. Many experiments prove that the method can improve detection and recognition performances for multi-classes space objects at different exploration distances.