动态视频序列下的坦克车体姿态识别

Tank Hull Pose Recognition in Dynamic Video Sequence

敌方坦克的姿态信息是装甲分队武器目标分配的重要依据,但目前坦克火控系统中并没有姿态识别的模块,为此提出一种基于压缩跟踪和级联分类器的动态视频序列下坦克车体姿态识别方法;将压缩跟踪算法的输出作为级联分类器的输入,缩小姿态识别的范围并减小分类器对不同环境中负样本的依赖;通过实验测试,将坦克车体姿态分为12组,训练了12个分类器,并在训练中引入了车体的轮廓信息;将多个级联分类器串联工作,识别坦克车体的不同姿态;对比了检测不同姿态的分类器识别效果,结果表明级联分类器能以一定的精度对在真实环境中的坦克姿态进行识别。

The posture of tank is important basis of armored unit weapon target assignment and there is not a pose recognition module in tank fire control system.So a method based on compressive tracking and cascade classifier that recognize the posture of tank body in a video sequence is proposed.The output of Compressive Tracking is used as the input of cascade classifiers which reduce the area that the cascade deal with and reduce dependence on negative samples of the cascade;The pose of tank hull is divided into 12 parts,12classifiers are trained and the contour information is introduced in the process of training;the cascades are cascaded to achieve the goal of recognizing the pose of tank hull.The effectiveness of the different cascade classifiers are tested and contrasted.The result illustrate the posture of tank body can be recognized by the cascade with certain precision.