Footprint Based Person's Weight Prediction with Dual Feature Extraction and Multi-Head Attention Mechanism

Footprints refer to traces formed by the interaction between the feet and the ground or other supporting surfaces during activities such as standing or walking, due to the weight of the body and the force of muscle contraction. This paper mainly focuses on the automatic prediction of body weight based on footwear footprints, specifically shoeprints. Firstly, a grayscale shoeprint image dataset is collected and constructed and then transformed into a pseudo-color shoeprint image dataset through pseudo-color transformation. Next, a dual feature extraction module composed of the GoogeLeNet and ResNet50 networks is employed to extract features, and a multi-head attention mechanism is adopted to weigh and fuse the input feature maps. Finally, the output features are processed through two fully connected layers to achieve body weight prediction for shoeprint images. Experimental results show that the proposed algorithm outperforms the state of art algorithms, and the accuracy rates for errors within 3 kg, 5 kg,7kg,8kg, and 10 kg are 36.22%,56.22%,77.30%, 85.41 % respectively.