Partial Similarity Human Motion Retrieval Based on Relative Geometry Features

Chen, Songle; Sun, Zhengxing; Li, Yi; Li, Qian

doi:10.1109/icdh.2012.91

Cited by 5 publications

(11 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the research of relevance feedback for CBMR is still at the outset stage. As far as the authors know, heuristic weight adjustments based on AnDCG [4] and based on RVII [5] are only two approaches in CBMR. It is necessary for CBMR to employ effective online learning mechanisms to further refine the retrieval performance.…”

Section: Related Workmentioning

confidence: 99%

“…Next, K features with the highest weights are inputted into the similarity calculation module to calculate the similarity between the query and each action in the database. In this paper, DTW algorithm [26] which widely used in CBMR [2,5] is used as the similarity measurement. Finally, top-N candidate actions with the lowest dissimilarity values are returned to the user.…”

Section: System Overviewmentioning

confidence: 99%

“…Tang et al [4] suggested adjusting the weight of each feature in the similarity measurement via the average normalized discounted cumulative gain (AnDCG), while Chen et al [5] proposed to set the weight of each feature via the ratio of the variance of interclass to intraclass (RVII). Both of them belong to heuristic methods.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Relevance feedback for human motion retrieval using a boosting approach

Chen

Sun

Zhang

et al. 2014

Multimed Tools Appl

Self Cite

View full text Add to dashboard Cite

Content-based human motion retrieval (CBMR) has been more and more important with the rapid growth of motion capture data, but the gap between high-level semantic concepts and low-level features hinders further performance improvement. Relevance feedback is an effective tool to narrow the semantic gap and enhance the retrieval performance. However, as a type of variable-length multivariate time series (VLMTS), motion capture data has its own characteristics including high-dimensionality, demand for elastic matching, and difficulty representing different movements in a uniform feature space, which make it much more challenging to design an effective relevance feedback approach. This paper presents a novel boosting approach for CBMR and the main contributions include three aspects. First, to fit in with the characteristics of VLMTS data and meet the real-time requirement of relevance feedback, the ensemble learning framework RankBoost is introduced and k-nearest neighbors combining with dynamic time warping (KNN-DTW) is employed as its weak ranker. Second, the set of extended Boolean geometry features containing much richer geometry elements and measures is used to represent motion content, and it provides a comparatively complete feature set for designing the weak ranker of RankBoost. Third, to solve the over-fitting problem caused by the small-sample training of relevance feedback, a novel learning objective composed of minimizing empirical ranking loss and minimizing the maximum generalization loss is proposed for RankBoost ensemble learning. Experimental results on CMU database and its extended database verify the effectiveness of the proposed approach.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: System Overviewmentioning

confidence: 99%

See 1 more Smart Citation

Relevance feedback for human motion retrieval using a boosting approach

Chen

Sun

Zhang

et al. 2014

Multimed Tools Appl

Self Cite

View full text Add to dashboard Cite

show abstract

“…The feature is improved to retrieve logically relevant motions by [28]. Those geometric features are also applied and extended by [29], where a combined relational geometric feature of over 20,000 dimensions is used. Since the feature is high dimensional, feature selection methods such as Adaboost are usually needed.…”

Section: Related Workmentioning

confidence: 99%

“…The dynamic time warping (DTW) algorithm is most widely used here, due to its effectiveness [8,29,30]. It is a dynamic programming (DP)-based algorithm that finds a path with a minimal distance between two motions.…”

Section: Related Workmentioning

confidence: 99%

A human motion feature based on semi-supervised learning of GMM

et al. 2014

View full text Add to dashboard Cite

those motion data to create realistic motions for new characters by applying motion editing [1,2], motion synthesis [3][4][5] and motion retargeting [6] techniques has become the focus of research in the past several years. However, prior to reusing and processing the old motions, one fundamental problem of identifying and extracting similar motion clips from the database has to be solved. It is essentially a motion matching problem. The general procedure is to calculate a concise and representative feature for each motion, and then compare the similarities with all other motions in the mocap database. The efficiency and accuracy of these motion retrieval and annotation processes largely depend on the property of the used features.Most early work [7] uses textual description such as running, walking, to label existing motions in a database. It does not only involve a lot of manual work, the textual label is also too short and general to fully represent the features of each motion. Later works [8][9][10][11] use the numeric-based features and logic-based features involving the 3D coordinates of each joint in each frame. It includes too much redundant information and the 'huge' feature makes the motion matching really slow. Some recent works [12] present semantic features which better represent the essence of motions and the low dimension of features largely speeds up the motion retrieval process. In this paper, we present a new feature in this category. The work [13] has shown that a human motion clip could be described with some representative poses, which we call the 'key-pose'. Intuitively, two similar motions may share most key-poses, while the motions belonging to different motion classes may share none or only a few key-poses (as shown in Fig. 1). A good selection of key-poses can be used to represent different motion classes. The second benefit to use key-poses as a feature is, although the category of motions is infinite, the types of key-poses are relatively limited. A new category Abstract Using motion capture to create naturally looking motion sequences for virtual character animation has become a standard procedure in the games and visual effects industry. With the fast growth of motion data, the task of automatically annotating new motions is gaining an importance. In this paper, we present a novel statistic feature to represent each motion according to the pre-labeled categories of key-poses. A probabilistic model is trained with semi-supervised learning of the Gaussian mixture model (GMM). Each pose in a given motion could then be described by a feature vector of a series of probabilities by GMM. A motion feature descriptor is proposed based on the statistics of all pose features. The experimental results and comparison with existing work show that our method performs more accurately and efficiently in motion retrieval and annotation.

show abstract

3D Human Motion Retrieval Based on Graph Model

Zhou

et al. 2019

E-Learning and Games

View full text Add to dashboard Cite

Partial Similarity Human Motion Retrieval Based on Relative Geometry Features

Cited by 5 publications

References 20 publications

Relevance feedback for human motion retrieval using a boosting approach

Relevance feedback for human motion retrieval using a boosting approach

A human motion feature based on semi-supervised learning of GMM

3D Human Motion Retrieval Based on Graph Model

Contact Info

Product

Resources

About