摘要
为解决汽车碰撞实验过程进行测量和记录数据困难的问题,提出一种能从汽车碰撞动态图像中检测、识别和追踪标志目标的算法。该算法采用二值图像同或相关法分离出目标和背景;在找出感兴趣区域后提取相邻帧目标的坐标和纹理进行特征匹配;给出了对匹配量化值进行加权平均融合的策略,并由等错误率最小准则确定融合的最佳权系数;在融合量化值定义为相似度的基础上,通过决策阈值对相邻帧目标间的最大相似度组合进行识别;从而进一步提出了用同构映射原则来判断相邻帧目标的最佳配对。实验结果表明,该算法对相邻帧目标的配对准确率比传统单特征法提高5%,能更有效的对目标进行追踪。
In order to solve the difficult problem in car crash experiment to measure and record data, a dynamic image from a car crash detection, target identification and tracking algorithm is proposed. The algorithm uses bi- nary images related to the target and background separation in the region of interest identified targets adjacent frames after the extraction of texture coordinates and feature matching; gives quantitative values of the matching strategy of integration of the weighted average by the criterion of minimizing error rate, etc. fusing the best weights; in the integration of quantitative value is defined as the basis of similarity, by decision threshold be- tween the objectives of the adjacent frames to identify the greatest similarity combinations; which further raised with the same structure mapping principle to determine the best matching target adjacent frames. Experimental results show that the algorithm matching adjacent frames target accuracy than the traditional 5% increase in sin- gle-feature methods, to more effectively track the target.
出处
《吉林大学学报（信息科学版）》
CAS
2011年第2期110-115,共6页
Journal of Jilin University（Information Science Edition）
基金
吉林省科技发展计划应用基础类研究基金资助项目(20090505)
关键词
汽车碰撞
目标追踪
同或相关
特征融合
同构映射
vehicle impact
object tracking
nclusive-OR correlation
feature fusion
isomorphic mapping
