基于改进HOG特征的空间非合作目标检测

传统的非合作目标检测方法大都基于一定的匹配模板,这不仅需要预先指定先验信息,进而设计合适的检测模板,而且同一模板只能对具有相似形状的目标进行检测,不易直接用于检测形状未知的非合作目标。为降低检测过程中对目标形状等先验信息的要求,借鉴基于规范化梯度的物体区域估计方法,提出一种基于改进方向梯度直方图特征的目标检测方法,首先构建包含有自然图像和目标图像的训练数据集;然后提取标记区域的改进方向梯度直方图特征,以更好地保持局部特征的结构性,并根据级联支持向量机训练模型,从数据集中自动学习目标物体的判别特征;最后,将训练后的模型用于检测测试集图像中的目标。实验结果表明,算法在由4953幅和100幅图像构成的测试集中分别取得94.5%和94.2%的检测率,平均每幅图像的检测时间约为0.031 s,具有较低的时间开销,且对目标的旋转及光照变化具有一定的鲁棒性。

Space non-cooperative target detection based on improved features of histogram of oriented gradient

Traditional non-cooperative target detection methods are mostly based on different matching templates which are well-designed with additional prior information. Moreover, one single template can be merely used to detect objects with similar shapes and structures, causing low applicability in detecting non-cooperative targets whose prior information are usually unknown. In order to solve those problems and inspired by the object estimation technique based on normed gradient, an object detection algorithm using improved features of histogram of oriented gradient is proposed. A training data set composed of natural images and target images is first built manually. Secondly, we extract the modified HOG information in the labeled regions to preserve detailed structures of the local features. Then, the cascaded support vector machine is used to train the model autonomously, which does not require prior information. Finally, we design several tests using the trained model to detect targets from the testing images. Numerous experiments demonstrate that the detection rates of the proposed method are 94.5% and 94.2% respectively when applied to testing sets with 4 953 and 100 images. The time consumption of extracting one image is about 0.031 s while it is robust to object rotation and illumination under certain condition.