基于LCSCA特征与协同表示的轨迹分析算法

Trajectory Analysis Algorithm Based on Least-squares Cubic Spline Curves Approximation Feature and Collaborative Representation

为了充分利用最小二乘三次样条近似(LCSCA)特征进行人体行为识别,提出一种基于协同表示分类(CRC)的轨迹分析算法;该算法将传统稀疏表示的分类识别算法中基于l_1范数的稀疏求解,改为基于l_2范数的协同表示求解,大幅降低了分类算法的复杂度;该方法将LCSCA轨迹特征与CRC分类器紧密结合,采用距离加权的Tikhonov矩阵增强分类效果。结果表明,该方法对于轨迹的等变化具备较强鲁棒性,算法运行速度较快。

Concerning the full use of least-squares cubic spline curves approximation （LCSCA） feature for human behavior recognition, the trajectory analysis algorithm based on collaborative representation classification （CRC） was proposed. This algorithm reduced the complexity of the classification algorithm by replacing the sparse representation classification based on 11 -norm with the collaborative representation classification based on l2-norm. LCSCA was combined with the CRC classifier to enhance the classification effect by using the distance weighted Tikhonov matrix. The results indicate that this method is robust to the change of trajectory and the operating velocity of this algorithm is faster.