Document Type



This item is available under a Creative Commons License for non-commercial use only


Computer Sciences

Publication Details

A dissertation submitted in partial fulfilment of the requirements of Technological University Dublin for the degree of M.Sc. in Computer Science (Data Analytics)


This study has investigated the potential application of machine learning for video analysis, with a view to creating a system which can determine a person’s hand laterality (handedness) from the way that they walk (their gait). To this end, the convolutional neural network model VGG16 underwent transfer learning in order to classify videos under two ‘activities’: “walking left-handed” and “walking right-handed”. This saw varying degrees of success across five transfer learning trained models: Everything – the entire dataset; FiftyFifty – the dataset with enough right-handed samples removed to produce a set with parity between activities; Female – only the female samples; Male – only the male samples; Uninjured – samples declaring no injury within the last year. The initial phase of this study involved a data collection scheme, as a suitable, pre-existing dataset could not be found to be available. This data collection resulted in 45 participants (7 left-handed, and 38 right-handed. 0 identified as ambidextrous), which resulted in 180 sample videos for use in transfer learning and testing the five produced models. The video samples were recorded to obtain the volunteers’ walking pattern, head to toe, in profile rather than head on. This was to allow the models to obtain as much information about arm and leg movement as possible when it came to analysis.

The findings of this study showed that accurate models could be produced. However, this varied substantially depending on the specific sub-dataset selected. Using the entire dataset was found to present the least accuracy (as well as the subset which removed any volunteers reporting injury within the last year). This resulted in a system which would classify all samples as ‘Right’. In contrast the models produced observing the female volunteers (the gender which also provided the highest number of left-handed data samples) was consistently accurate, with a mean accuracy of 75.44%. The course of this study has shown that training such a model to give an accurate result is possible, yet difficult to achieve with such a small sample size containing such a iii small population of left-handed individuals. From the results obtained, it appears that a population has a requirement of >~21% being left-handed in order to begin to see accuracy in laterality determination. These limited successes have shown that there is promise to be found in such a study. Although a larger, more wide-spread undertaking would be necessary to definitively show this.