Evaluation of Genetic Programs in Multiple Cases for Gait Classification and Recognition

Abstract

The analysis of human motion is a challenging research domain that attracts the attention of researchers from several disciplines, including sociopsychology, neurobiology, and computer science. A successful recognition of the person's walk could be used for personal identification, and also, would be important for understanding the human's emotions, personality, and neurological disorders. However, recognizing the human gaits is a challenging task because of the complexity of the eventual analytical model that defines the numerical relationship between the relevant features of the gait. In our previous work we proposed an approach of applying genetic programming to automatically design such a model in a way much similar to the evolution in nature. In this paper, we continue the focus on human gait recognition, and present an analysis of the trade-off between the evolution of genetic programs (GPs) and their performance. We consider different training cases, provided that the computational resources and other parameters are kept constant. Furthermore, in our previous work, there was an important unanswered question regarding the effect of the increased number of fitness cases and the use of experts in collaborative filtering on the evolution of GPs and gait recognition. This study is an attempt to explore the same unexplained question.