基于评价向量的异源图像目标检测  

Objects Detection in Different Source Images Based on Evaluation Vector

作　　者：郗润平[1,2] 贾高云 张艳宁 张福俊[1,2] XI Runping;JIA Gaoyun;ZHANG Yanning;ZHANG Fujun(School of Computer Science,Northwestern Polytechnical University,Xi’an 710129,China;Shaanxi Provincial Key Laboratory of Speech&Image Information Processing,Xi’an 710129,China)

机构地区：[1]西北工业大学计算机学院,西安710129 [2]陕西省语音与图像信息处理重点实验室,西安710129

出　　处：《计算机工程与应用》2019年第1期180-185,270,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61572405;No.61231016;No.61303123);国家高技术研究发展计划(863)(No.2015AA016402);模式识别国家重点实验室开放课题(No.201600038)

摘　　要：在异源图像运动目标检测中,对不同源信息处理的可信度量是影响异源协同检测的关键。针对传统单源目标检测中漏检率、误检率高等问题,提出了基于评价向量的异源图像目标检测方法。通过引入目标面积检测惯性、目标数量检测惯性和目标独立完整性三个评价因子,构造出用来评价不同信息源运动检测结果好坏的评价向量,并运用改进的k-means聚类算法产生目标中心向量,最后利用协作与竞争机制对聚类相似度进行反馈,实现了多源图像的协同检测。实验结果表明,相比于传统的单源检测算法和融合检测算法,该算法具有较高的检测精度和较低的漏检率、误检率。How to evaluate the reliabilities of different image sensors and their processing result is an important issue in the field of multi-modal objects detection. As single-source sensor in object detection has the disadvantages of high miss rate and mistake rate, this paper proposes a new approach, which can evaluate the reliabilities of the detection results in different source images. Firstly, three evaluation factors of inertia, target number of inertia and target independent integrity are introduced to construct an evaluation vector to assess the quality of motion detection. Secondly, k-means clustering are used to generate the target center vector. Then the cooperative and competitive mechanism are applied to feedback the clustering similarity. Finally, the objects detection in different source images is realized. Simulations on lots of images verify that the new proposed approach is more robust for lower detection error rate and false-negative rate.

关 键 词：异源图像 目标检测 评价向量 K-MEANS聚类 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]