基于图像信息的多特征空间坦克姿态估计

摘要

Tank as the main tatget of ground,it is important to analyze its pose.Considering the need of tank pose estimation,collecting different pose images as training images,principal component analysis is being used to calculated feature vectors,it turns out that each image set can be represented as three feature vectors,so few feature vector to establish eight feature spaces is used,obviously reduces the dimension of feature space.Based on the cosine value between object vector and its reconstructed vector,tank’s spatital position can be located,finished the tasks of pose estimation.Simulations results show that using this method recognize the tank pose estimation.%坦克作为地面战场的主要目标，分析其姿态至关重要。根据坦克姿态估计的需要，在可见光条件下采集了坦克车体纵轴与瞄准镜光轴不同夹角的图像作为训练集。利用主成分分析法选取了目标的主要特征向量，每个训练子集用3个特征向量表示，利用少量的特征向量建立目标的8个特征空间，降低了空间的维数。设计判别准则将待识别目标向量与重构向量之间的余弦值进行比较，即确定目标所在的空间位置，完成了目标姿态的识别。实验结果表明，利用建立目标多特征空间的方法识别目标空间位置是有效的。