A study on gait analysis using moving silhouettes: Striking a balance with simplicity between feature generalization and specificity

Feature selection has been identified as one of the most effective ways to differentiate patterns. Feature generalization and specificity are two major challenging problems that are hard to satisfy at the same time. This study aims to propose a method that makes an attempt to retain both feature generalization and specificity. Simplicity is another factor that we take into account. The proposed model is applied to extract features so as to separate different walkers in motion. Feature selection is performed by extracting silhouettes. Two sets of features, i.e. temporal features and flexibilities, are tested and then eliminated to improve the generalization capability. Experimental result shows that the proposed method achieves a satisfactory result in separating different walkers and tolerating noises. More importantly, it demonstrates well in striking a balance between generalization and specificity with satisfactory simplicity.