微型电脑应用

智能音视频分析系统中的声源运动目标检测与识别是通过估计多快拍的声源方位角实现的。为了在多快拍的前提下尽量发挥阵列行信号的稀疏先验作用，提出基于自适应套索的稀疏贝叶斯学习(aLASSO-SBL)方案。但由于该方案框架的参数更新需要复杂的矩阵运算，因此提出空间轮换变元(SRV)的优化方案，结合声源独立的特性轮换进行变量的更新，从而降低算法的运算复杂度。仿真实验结果表明，在信噪比为0 dB～20 dB和连续快拍数量小于5的条件下，所提算法具有明显的优势，能够更加准确地识别声源方位角。

The detection and recognition of sound source moving target in intelligent audio and video analysis system is realized by estimating the sound source azimuth of multiple snapshots. In order to maximize the sparse prior function of array row signals under the premise of multiple snapshots, an adaptive LASSO based sparse Bayesian learning(aLASSO-SBL) scheme is proposed. However, since the parameter update of the scheme framework requires complex matrix operations, an optimization scheme of space rotation variable(SRV) is proposed, and the variables are updated in combination with the independent characteristics of the sound source, thereby reducing the computational complexity of the algorithm. The simulation results show that the proposed algorithm has obvious advantages under the conditions of signal-to-noise ratio of 0 dB～20 dB and the number of snapshots less than 5. It can identify the sound source azimuth more accurately.