4th International Conference on Computer Science and Engineering, UBMK 2019, Samsun, Turkey, 11 - 15 September 2019, pp.394-397
The effective use of biometric features which are unique and measurable leads the way in successful analyzing of the characteristics of human biometry. One of these developing biometric technologies is gait analysis. Systematic review of gait and unrevealing of the correlation between body parts helps to find out meaningful information quite easily. In this study, we utilized a Convolutional Neural Network (ConvNet/CNN) which can take in gait data, assign importance (learnable weights and biases) to various aspects/objects and helps to differentiate one from another. Thus, we could detect gender types by using some distinguishing gait features. The importance of using light version of CNN is about efficiency in computational costs and storage spaces. A variation of maxout activation, called max-feature-map (MFM), into each convolutional layer of CNN, is the first key factor in our study. After this step, we focused on designing four alternative networks to obtain successful results. All experiments have been conducted on three-dimensional gait datasets which are obtained by using reflectors all over the body. Three-axis gait dataset with the learned single network achieves state-of-the-art results on gender detection.