Subjective Baggage-Weight Estimation Based on Human Walking Behavior

Abstract

We address a new computer vision problem of subjective baggage-weight estimation, where the term subjective weight is defined as how heavy the person feels. In this paper, we propose a method named G2SW+ (Gait to Subjective Weight plus), which is an extension of our previous method, G2SW. The method uses human walking behavior, including 3D locations and velocities of body joints and silhouettes, as input. It estimates the subjective weight using a combination of a Convolutional Neural Network and a Graph Convolutional Network. It also estimates human body weight and recognizes the type of baggage as subtasks based on the assumption that body weight and type of baggage affect human gait. For the evaluation, we built a dataset for subjective baggage-weight estimation, consisting of pairs of 3D skeleton and human silhouette sequences with subjective weight, body weight, and baggage-type annotations. We confirmed that the proposed method can accurately estimate the subjective baggage weight. Moreover, we confirmed that training with the subtasks and utilizing the human silhouette sequence as an additional input improves the performance of the subjective weight estimation.

Keywords

• Subjective baggage-weight
• gait to subjective weight plus (G2SW+)
• human silhouette image
• graph convolution
• multi-task learning