一种监督降维的OP-LASRC算法在行为识别中的应用

发布时间：2018-02-02 10:13

本文关键词： 稀疏表示监督降维优化投影线性近似行为识别　出处：《计算机应用研究》2017年11期 　论文类型：期刊论文

【摘要】：针对大数据的人体行为识别时实时性差和识别率低的问题,提出了优化投影对线性近似稀疏表示分类(OP-LASRC)的监督降维算法,与线性近似稀疏表示(LASRC)快速分类算法结合用于大数据的行为识别。利用LASRC的残差计算规律设计OP-LASRC实现监督降维:在追求一个线性投影时减小训练样本的本类重构残差及增大类间重构残差,保留样本的类别特征。对降维后的行为数据用LASRC算法分类:用L2范数估算稀疏系数,取前k个最大的稀疏系数对应的训练样本,用L1范数和残差计算得到识别结果。在KTH行为数据库上的实验表明,OP-LASRC降维后,LASRC在分类时识别率高达96.5%,执行时间比同类算法短,抗噪声能力强,证明了OP-LASRC的高效和强鲁棒性,能完美匹配LASRC用于大数据的行为识别。
[Abstract]:In order to solve the problem of poor real-time and low recognition rate in big data's human behavior recognition, a supervised dimensionality reduction algorithm for linear approximate sparse representation classification (OP-LASRC) based on optimal projection is proposed. LASRC) and linear approximate sparse representation. The fast classification algorithm is combined with big data's behavior recognition, and the OP-LASRC is designed to reduce the dimension by using the residual calculation rule of LASRC. In pursuit of a linear projection, the class reconstruction residuals of the training samples are reduced and the inter-class reconstruction residuals are increased. LASRC algorithm is used to classify the behavior data after dimensionality reduction: the sparse coefficients are estimated by L2 norm, and the training samples corresponding to the first k largest sparse coefficients are selected. The recognition results are obtained by using L1 norm and residuals. Experiments on KTH behavioral database show that the recognition rate of OP-LASRC in classification is as high as 96.5%. The execution time is shorter and the anti-noise ability is stronger than the similar algorithms. It is proved that OP-LASRC is efficient and robust, and it can perfectly match LASRC for big data's behavior recognition.
【作者单位】：上海理工大学光电学院;上海市现代光学系统重点实验室;
【基金】：国家自然科学基金资助项目(41075019)
【分类号】：TP391.41
【正文快照】： 0引言人体行为识别是模式识别和机器视觉领域中被广泛关注的研究热点,不仅在智能监控、运动分析、身份鉴别和人机交互方面具有广阔的应用前景,而且在安全隐患场所的异常行为监控,如交通事故、用电安全、医疗监护等方面具有重要的意义[1]。人体目标的行为动作在空间上表现出复

【相似文献】