Segmenting motions and detecting primitive movements involve extracting common patterns among different actions, leading to estimates of individual activity possibilities. Human motion data generally contains temporal context, and entire body movements are represented by high-dimensional data structures. Consequently, the temporal correlations between elements in time-series data are complex, making the detection of segments from such data sequences a challenging problem. In this study, we propose a human motion segmentation method based on a topological clustering approach that considers temporal relationships, utilizing the learning algorithm of growing neural gas (GNG). While GNG can dynamically adjust its topological structure according to the probability density of data, it does not include an algorithm to extract the temporal relationships between nodes. Therefore, we propose a topological clustering approach for human motion segmentation based on GNG with temporal coding. The effectiveness of the proposed method is investigated by comparing it with standard GNG using several test datasets and an actual motion dataset.