Remarks on human posture classification using self-organizing map

Estimation of pointing poses for visually instructing mobile robots under real world conditions

“…Such a silhouettebased approach is employed by Rogalla et al [3], Urano et al [15] and Takahashi and Tanigawa [16]. In Section 5 we give an overview of the results we achieved with this relatively simple method of feature extraction and compare it with the results of the Gabor filter-based one.…”

“…A similar system with one function approximator (a Multi-Layer Perceptron-MLP) is used by Stiefelhagen [17] to estimate the line of sight of a user. Certainly other authors use other neural networks as approximators: Krueger and Sommer [18] employ a Local Linear Map (LLM) to estimate the head-position of a person, Takahashi and Tanigawa [16] a Self-Organizing Map (SOM) to classify different poses, and Paquin and Cohen [4] compare the extracted features with labeled data with a method very similar to a k-Nearest Neighborhood method. We implemented those classifiers, that have been successfully used in the different aforementioned approaches and compared the results with our function approximator presented in [1,2] to determine the best method for this target position estimation task.…”

“…A second feature extraction we used is the histogram of the user-silhouette as proposed by Takahashi and Tanigawa in [16]. This method also uses a background subtraction to separate the user from the background.…”

“…Feature extraction as proposed by Takahashi and Tanigawa in[16]. The number of human-pixels in each line and column is used as feature for the approximation of the target.value if the samples of each class have a little intra-class variance (the denominator) and if the different classes do not overlap (the inter-class variance given in the numerator).…”

Self Organizing Maps for Reducing the Number of Clusters by One on Simplex Subspaces

“…A number of 5384 papers related to the SOM are listed in [2,3] and cited at www.cis.hut.fi/research/ som-bibl/. Recent applications of the SOM include the classification of human body postures from images [4], the grouping and visualizing human endogenous retroviruses [5], speaker clustering [6], to mention a few.…”