The sparse collaborative tracking (SCT) method has been developed for object tracking recently, and it is very efficient and robust to various occlusions. In SCT, sparse representation (SR) plays an essential role because it needs to perform several manipulations of sparse matrix representation (SMR) or nonnegative SMR in each iteration. So one of the most challenging problems in SCT is how to efficiently solve the SMR. However, existing SR algorithms are solely developed for the vectors-based SR. They partition SMR into a set of vector-based SR problems and solve them in the level-2 BLAS (Basic Linear Algebra Subprograms) manner, i.e., matrix-vector operations, which is computationally much less efficient than the direct level-3 BLAS (direct matrix-matrix operations). To solve this problem, by extending the standard SR algorithm from the vector version to the matrix for SMR and nonnegative SMR, BLAS3-based Sparse Learning (BLAS3-SpaL) is first developed, and then the corresponding BLAS3-SpaL-based SCT method (FastSCT-BLAS3SpaL) is further developed for fast robust-object-tracking in this paper. The experiments verified that it achieves robust object tracking by reducing accumulation errors and speeds up tracking with more than double speed.

Abstract Despite their success, existing human pose estimation approaches mostly have complex architectures, high cost, and lack of lightweight modules. To address this problem, this paper proposes a Ghost Shuffle Lightweight Pose Network (GSLPN) with a more lightweight and efficient network architecture than the popular Lightweight Pose Network. First, in order to condense the scale of the network while maintaining its performance, we stack two lightweight modules, depthwise convolution and the Ghost module, to build our initial prototype bottleneck. Then, we impose a channel shuffle operation on the prototype bottleneck to shuffle the sequence of the feature maps for constructing Ghost Shuffle Bottleneck (Ghost Shuffle Bottleneck) with effective feature representation so as to develop a GSLPN. Second, a lightweight, efficient parallel attention mechanism, Lightweight Pose Parallel Attention, is proposed to improve keypoint locating accuracy. An experiment validating the proposed method showed that GSLPN achieved competitive performance with a smaller model size and less computational complexity than state‐of‐the‐art methods, indicating that the GSLPN is a superior approach for human pose estimation.

A sparse partial update (SPU) algorithm and its improved version improved SPU (ISPU) algorithm, are proposed in this paper for sparse system identification. The SPU first categorizes its filter coefficients into active and inactive coefficients. Then all the active coefficients are included in each adaptation, while only a small portion of the inactive coefficients is periodically chosen to be included in the adaptation. The SPU emphasizes convergence of the active coefficients by updating them at every adaptation, while the periodical adaptation of the inactive coefficients ensures its robustness and tracking capability. By eliminating most of the inactive coefficients from adaptation, the SPU significantly reduces the number of adapting coefficients in each adaptation. The decline in the number of adapting coefficients eventually leads to improvement in both computation and convergence speed. To avoid performance degradation in the case of identifying a dispersive system, an ISPU is further proposed by making modifications to the SPU. The simulation results demonstrate that the ISPU not only outperforms other sparse adaptive algorithms in identifying a sparse system but also performs at least as well as the NLMS algorithm in identifying a dispersive system.

A novel scheme is proposed to locate the dispersive region, whose location is essential for parallel echo cancellation. In the scheme, a first filter adapts to a subsampled version of the input signal to roughly identify the impulse response. After each adaptation, a squaring function and a moving window integration procedure are performed on the first filter, and the region with the maximum integration value is considered to be the dispersive region. Finally, a second short filter is used to precisely identify the active coefficients belonging to the located dispersive region to implement the actual echo cancellation. Simulation results suggest that the parallel structure improves its convergence speed 3 times and its computation can be reduced by 3/8 compared with the traditional NLMS algorithm by decreasing the filter length. Due to the more accurate estimate of the location, the misalignment noise of the proposed algorithm is at least 10 dB lower than that of the conventional dispersive region locating algorithm. Moreover, the proposed algorithm based on the parallel structure outperforms other sparse adaptive algorithms in all aspects.