摘要
视觉特征提取质量决定了UCAV认知导航的效能。为实现UCAV认知导航的高鲁棒性特征点提取,提出一种特征点优选的多元量化MQ-SIFT方法。针对SIFT模拟特征序列分布不均匀、正确匹配率不高的缺陷,提出采用多值量化与比特抽取结合法对模拟特征序列进行多元量化,并且分析验证了该方法的优越性能。为确保高鲁棒性特征点用于认知导航,对特征点进行了优选,给出了优选准则,提出了搜索最大连通集的改进迭代互欧氏距离方法。仿真结果表明:在图像信噪比大于10 dB时,MQ-SIFT算法及其优选的特征点具有较高的正确匹配率,并且其匹配率能够满足认知导航系统需求。
The extraction of vision key- points decides directly cognitive navigation efficiency for UCAV. In order to extract high robust key-points for UCAV, an algorithm named multi -quantifying scale invariant feature trans-form （ MQ - SIFT） with key - points optimization is proposed. According to the deficiency of SIFT algorithm in an-alogue feature vectors＇balance and correct matching score, a method combining the multiple value quantifying and reshaping operation is presented to quantify analogue feature vectors. The analysis and simulation results verify the better properties of this method. Furthermore, in order to perfect the property of MQ - SIFT with fewer robust key - points, the optimization rules are discussed, and an iterative cross-Euclidean distance search method is proposed to search the maximum connected set. Simulation results show that MQ - SIFT algorithm has higher correct matc- hing score with signal-to-noise （SNR） above 10 dB, and their matching score can meet the requirements of cog-nitive navigation system.
出处
《空军工程大学学报（自然科学版）》
CSCD
北大核心
2012年第4期65-69,共5页
Journal of Air Force Engineering University(Natural Science Edition)
基金
国家自然科学基金资助项目(61001111
61104056)
