战场可疑目标智能识别与跟踪框架研究

Research on Framework of the Suspicious Targets Recognition and Tracking in Battlefield

针对陆战场人、装备等典型目标图像,提出了一套战场可疑目标智能识别与跟踪框架,用于目标检测与图像识别,以及对可疑目标进行跟踪。人脸检测基于MTCNN网络,可以有效检测人脸及人脸中的5个关键点;人脸识别基于FaceNet网络,并且使用随机森林算法作为分类器,训练用于识别10位在影视剧中饰演军人的演员模型,在测试集上的准确率为94.76%;装备目标检测与识别基于YOLO v3模型,能够检测与识别20种武器装备;可疑目标跟踪基于BACF算法,在10个10 s测试视频中,跟踪准确率为90%,平均输出速率为24 fps。

Aimed at the typical target images of people and equipment in land battlefield,it was proposed that a set of intelligent recognition and tracking framework for the suspicious targets used for the target detection and image recognition,as well as tracking suspicious targets.Face detection was based on MTCNN,and could effectively detect five key points in face.Face recognition was based on the FaceNet,and took the random forest algorithm as classifier.The model was trained on using ten soldier players in the film and television drama.The model had an accuracy rate of 94.76%on the test set.The equipment target detection and recognition was based on the YOLO v3,which could detect and identify 20 kinds of equipment.The suspicious target tracking was based on the BACF algorithm.For 10 testing videos of 10 seconds,the tracking accuracy was 90%and the output rate was 24 fps.