A novel method to predict slip resistance of winter footwear using a convolutional neural network

Abstract

Slips and falls on ice are common causes of injury in the winter and can result in economic loss. Using slip resistant winter footwear is a key factor in reducing the risk of slips and eventually falls. In this study, we develop a model that classifies footwear outsoles based on how slip resistant they are on icy surfaces. Our dataset consisted of outsole images of 89 winter footwear samples that were previously tested and rated with a human-centred protocol called the Maximum Achievable Angle (MAA). We applied a transfer learning technique where the best classification model used the DenseNet169 pre-trained model and obtained an accuracy and F1-score of 0.93 ± 0.01 and 0.73 ± 0.03, respectively. Our results suggest that the proposed model was able to properly identify high and low slip resistance quality tread patterns. Our findings confirmed that a footwear’s tread pattern has a direct impact on its slip resistance. The proposed model will help footwear manufacturers to improve their workflow and increase product quality which can ultimately decrease the events of slips and falls.