Adaptive multi-view feature selection for human motion retrieval

Abstract

a b s t r a c tHuman motion retrieval plays an important role in many motion data based applications. In the past, many researchers tended to use a single type of visual feature as data representation. Because different visual feature describes different aspects about motion data, and they have dissimilar discriminative power with respect to one particular class of human motion, it led to poor retrieval performance. Thus, it would be beneficial to combine multiple visual features together for motion data representation. In this article, we present an Adaptive Multi-view Feature Selection (AMFS) method for human motion retrieval. Specifically, we first use a local linear regression model to automatically learn multiple view-based Laplacian graphs for preserving the local geometric structure of motion data. Then, these graphs are combined together with a non-negative view-weight vector to exploit the complementary information between different features. Finally, in order to discard the redundant and irrelevant feature components from the original highdimensional feature representation, we formulate the objective function of AMFS as a general trace ratio optimization problem, and design an effective algorithm to solve the corresponding optimization problem. Extensive experiments on two public human motion database, i.e., HDM05 and MSR Action3D, demonstrate the effectiveness of the proposed AMFS over the state-of-art methods for motion data retrieval. The scalability with large motion dataset, and insensitivity with the algorithm parameters, make our method can be widely used in real-world applications.

Retrieval of spatial–temporal motion topics from 3D skeleton data

“…Furthermore, the retrieval approach proposed by Zhu et al [51] highly relies on the manually designed geometric thresholds based on HDM05, which may lead to less robustness in other challenging dataset (e.g., Berkeley MHAD). Compared with Wang et al [40], our approach shows a more stable performance on all the motion types. The high-dimensional features considered in [40] claim to optimize more parameters with a longer retrieval time.…”

“…To validate our approach, we examine the deep learningbased motion image representation proposed by Du et al [8], the adaptive motion feature selection proposed by Wang et al [40], the temporal sparse representation-based motion retrieval proposed by Zhou et al [50], the topic modeling retrieval approach based on geometric motion features proposed by Zhu et al [51] and the benchmark retrieval algorithm based on semantic features proposed by Qi et al [30], and compare their performance with ours on the same evaluation datasets. All the methods are tested on equal terms of experimental protocol.…”

“…The temporal pyramid variation, JR with BPR descriptors generation and LDA with KLD (Kullback-Leibler divergence) are shared for both training and query processes joint classes, joint positions and temporal alignment. Wang et al [40] automatically select the most informative motion descriptors extracted from the well-defined trajectory-based [10] and pose-based features [6]. Hussein et al [12] utilize the covariance matrix as the spatial descriptor of joints to classify actions, and the covariance descriptor is later employed in [21] to construct adaptive graph kernels.…”

DeepHuMS: Deep Human Motion Signature for 3D Skeletal Sequences

“…Most approaches on the 3D human motion retrieval have focused on developing hand crafted features to represent the skeleton sequences [11,15,20]. In this section, we broadly categorize them by the method in which they engineer their descriptors.…”

Motus: Marco de trabajo para la clasificación de Captura de Movimiento humano en ambientes no controlados

“…Feature selection (see figure 2): Figure 2. Adaptive multi-view feature selection (AMFS) [11] Model Selection: Two-Step SVM Fusion: Combine tree-structured vector quantization and Multiple binary Support Vector Machine classifiers [3]. The proposed algorithms using the 5-fold cross validation procedure and obtained a correct classification rate of 99.6% [3].…”