|
Publications | Pose eval. techniques | Number of stages | Algorithm | DOF | Remarks |
|
Ivekovic and Trucco (2006) [22] | HP | 6 | PSO | 20 | The approach is only performed for upper body pose estimation |
Robertson and Trucco (2006) [21] | HP | 6 | Parallel PSO | 24 | The approach is only performed for upper body pose estimation |
John et al. (2010) [6] | HP | 12 | HPSO | 31 | The algorithm is unable to escape from local maxima which is calculated in the previous hierarchical levels. However, the approach is computationally more efficient than the competing techniques |
Ivekovic et al. (2010) [41] | HP | 12 | APSO | 31 | Computationally very inexpensive that is useful for real-time applications |
Mussi et al. (2010) [46] | HP | 11 | PSO | 32 | The approach has been implemented in GPU which significantly saves the computational cost when compared to sequential implementation |
John et al. (2010) [7] | HP | 12 | Manifold constrained PSO | 31 | Charting a nonlinear dimension reduction algorithm has been used to learn motion model in low-dimensional search space. However, the approach has not been tested with public available dataset |
Krzeszowski et al. (2010) [39] | Holistic | * | PF + PSO | 26 | Mobility limitation is imposed to the body model and is computationally expensive. Moreover, approach has not been tested with publicly available dataset |
Zhang et al. (2010) [40] | Holistic | * | APSOPF | 31 | The approach is able to alleviate the problem of inconsistency between the observation model and the true model |
Yan et al. (2010) [42] | Holistic | * | AGPSO | 29 | Computationally very expensive |
Kiran et al. (2010) [43] | — | — | PSO + K | — | The approach is only tested for posture classification |
Zhang et al. (2011) [11] | HP | 5 | PSO | 29 | The approach produces good tracking results but suffers from heavy computational cost due to more numbers of fitness evaluation |
Krzeszowski et al. (2011) [10] | HP | 2 | GLPSO | 26 | The approach suffers from error accumulation and mobility limitation is imposed to the body model. Moreover, approach was not tested with publicly available dataset |
Kwolek et al. (2011) [9] | HP | 2 | GLAPSO | 26 | The approach saves the computational cost by using 4 core processor computing powers. However, the approach was not tested with publicly available dataset |
Kwolek et al. (2011) [8] | Global | * | latency tolerant parallel PSO | 26 | The approach demonstrates the parallel nature of PSO and its strength in computational cost as compared to multiple PC (8) versus single PC (1) |
Kwolek (2011) [26] | HP | 3 | PSO | 26 | Tracking in surveillance videos which can contribute toward the view-invariant action recognition |
Kwolek et al. (2012) [23] | Holistic | * | RAPSO, APSO | 26 | The approach produced better results than PF and APF, but suffers from large computational cost |
Zhang and Seah (2011) [45] | HP | 5 | NFS, BBPSO | 36 | NFS algorithm produces good tracking results and their GPU implementation is able to give real-time results |
Ugolotti et al. (2013) [17] | HP | 11 | PSO, DE | 32 | Article demonstrates the strengths of two evolutionary approaches (PSO and DE) on GPU implementation |
Fleischmann et al. (2012) [12] | SP | 2 | SPPSO | 31 | The soft partitioning strategy overcomes the error accumulation issue. However, SPPSO suffers from heavy computation cost |
Li and Sun (2012) [50] | — | — | SAPSO | — | Principal component analysis (PCA) has been used to reduce the dimensionality and learn the latent space human motion which is the main novelty of the work |
Saini et al. (2012; 2013) [37, 38] | HP | 12 | QPSO | 31 | H-charting algorithm has been used to reduce the search space and learn the motion model. Proposed QPSO algorithm is able to escape from local minima |
Zhang et al. (2013) [48] | HP | 2 | NFS | 36 | New generative sampling algorithm with a refinement step of local optimization has been proposed. Moreover, the approach does not rely on prior strong motion. Due to GPU implementation approach, it is able to give real-time performance |
Nguyen et al. (2013) [51] | HP | 10 | HAPSOPF | — | The body pose is optimized hierarchical manner in order to reduce the computational cost of APSOPF |
Zhang (2014) [44] | Holistic | * | SAPSO | — | The main novelty of work is color, edge, and motion cue are integrated together to construct the weight function |
Rymut and Kwolek (2014) [49] | Holistic | * | PSO | 26 | Parallelization of the cost function is the main novelty of the work. Furthermore, CPU versus GPU performance has been demonstrated for human motion tracking |
|