基于LBFGS-EKF算法的三维空间目标跟踪研究

Research on target tracking based on limited-memory BFGS and extended kalman filter in 3D-space

人机交互中,人手与机器的交互是最通用的方式,因此手势交互研究是当今人机交互研究的重点之一。本文以Leap Motion为依据,对三维空间目标检测和跟踪进行研究,提出了一种限定内存扩展卡尔曼滤波算法(LBFGSEKF)。该算法在基于降低噪声所提出的EKF方法上,通过LBFGS算法求最优解的方式来代替EKF算法在每次迭代中求逆Hessian矩阵而造成内存消耗,计算速率下降,导致实时性差的问题,从而形成一种新的目标跟踪算法。仿真结果表明,新算法用于手势识别时,可以降低误差、提高目标跟踪的精度。

In human-computer interaction,the interaction between human-hands and machine is the most common way,so gesture interaction research is one of the focuses of human-computer interaction research today.Based on Leap Motion,this paper studies the three-dimensional space object detection and tracking,and proposes a new algorithmLBFGS-EKF algorithm.The algorithm is based on the EKF method proposed by reducing noise,through the method of the LBFGS algorithm is used to replace the Hessian matrix in each iteration instead of the EKF algorithm,which consumes memory and reduces the computational rate,resulting in poor real-time,thus forming a new target tracking algorithm.The simulation results show that by this new algorithm LBFGS-EKF for gesture recognition,the error can be reduced and the real-time performance of target tracking can be improved.