摘要
以海港集装箱扭锁自动化装卸过程中扭锁的认知识别为研究背景,采用Kinect传感器感知环境及物体信息,提出基于改进的增量主成分分析(Principal component analysis,PCA)方法构建实时认知识别系统。在系统在线学习阶段,基于新样本与已有特征重建样本之间的差异程度来感知、监测新类别输入,控制特征向量增量式更新;基于类内距离比较,优化特征向量组合,并自适应地更新类内距离阈值;将高维视觉信号转化为低维的机器人内部表达形式,从而在线实时地学习、更新、累积特征知识,同时完成模式识别任务。实验结果表明:该方法在有效提高视觉系统实时性、自适应性、稳定性及识别准确率的同时,控制了特征维度,从而减少了数据处理量及存储空间。
Research of the cognitive recognition of twist-lock automation handling system is conducted.In this research,Kinect is employed to collect environment and objects information,and an improved incremental Principal Component Analysis(PCA)is proposed to build real-time cognitive recognition system.In online learning phase,the new class is monitored and feature vectors are updated incrementally based on the difference between the new input and the reconstruction one using current eigenvectors;the feature vectors are optimized and the inner-class distance threshold is updated adaptively based on comparison of inner-class distance.Thereby,the proposed algorithm can convert high-dimension information to low-dimension machine expression,learn,update and accumulate feature knowledge online,and complete pattern recognition task at the same time.Experiment results show that the proposed algorithm can improve the adaptability,robustness,recognition rate and realtime performance of a visual system,Moreover,calculation and storage space can be reduced bycontrolling the feature space dimension.
出处
《吉林大学学报(工学版)》
EI
CAS
CSCD
北大核心
2016年第3期890-896,共7页
Journal of Jilin University:Engineering and Technology Edition
基金
新加坡创新基金项目(11-27801-36-R140)
新加坡教育部创新基金项目(MOE2013-TIF-1-G-057
MOE2013-TIF-2-G-040)
国家自然科学基金项目(51275065)
关键词
计算机应用
在线学习
增量主成分分析
自适应特征更新
感知识别
computer application
online learning
incremental principal component analysis(PCA)
adaptive feature update
cognitive recognition