论文部分内容阅读
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4 953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0.031s,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。
Traditional non-cooperative target detection methods are mostly based on a certain matching template, which requires not only prior identification of a priori information, and the design of a suitable detection template, and the same template can only detect targets with similar shapes and can not be directly used for detecting shapes Unknown non-cooperation goals. In order to reduce the requirement of prior information such as target shape in the detection process, a method of object detection based on gradient histogram feature is proposed by using the method of object region estimation based on normalized gradient. Firstly, a new method of object detection with natural image and target image Training the dataset; and then extracting the improved directional gradient histogram features in the marked region to better maintain the structural features of the local features and automatically learning the discriminant features of the target object from the data set according to the training model of the cascade SVM; and finally, The trained model is used to detect the target in the test set image. Experimental results show that the algorithm achieves detection rates of 94.5% and 94.2% respectively in the test set consisting of 4 953 images and 100 images, and the average detection time of each image is about 0.031s, which has a lower time cost. The rotation of the target and the illumination change have a certain robustness.