Visual gesture recognition is one of the main areas of research in human-computer and human-robot interfaces. In this paper we present a novel visual gesture recognition method for aircraft marshalling signals. To capture hand motion information, we used a color-based tracking algorithm with an adaptive window for each frame. A feature selection algorithm was used to classify the motion information into four different gesture phases. By using the gesture phase information, we built the gesture model, which consisted of a symbol sequence and a number of probabilities. Each gesture model was learned from the longest common subsequence (LCS) of a set of symbol sequences and the probability of the symbols. A similarity measure using the proposed gesture model is presented that combines the deterministic and probabilistic matching scores. In the experiments we show the efficiency and accuracy of the proposed method.