Analysis of Human Motion Data for Vehicle Ingress Discomfort Evaluation

Abstract

The ease of entering a vehicle, known as ingress, is one of the important ergonomic factors that car manufacturers consider during the process of vehicle design. This has motivated vehicle manufacturers to focus on assessing and improving ingress discomfort. With the rapid advancement in human motion capture and computer simulation technologies, one of the promising means to evaluate vehicle ingress discomfort is through analyzing human motion data. For this purpose, this dissertation will focus on proposing methods that analyze human motion data to evaluate vehicle ingress discomfort. The first part of this dissertation proposes a method for identifying and analyzing human motion variation patterns. The method uses a high-order array to represent human motion data and utilizes the Uncorrelated Multilinear Principal Component Analysis (UMPCA) method to identify variation patterns in human motion. The proposed method is capable of preserving the original spatiotemporal correlation structure of human motion data and provides better feature extraction than Principal Component Analysis (PCA). The method is applied to the ingress motion data to show its effectiveness in automatically detecting important motion variation patterns. The second part of this dissertation proposes a method for modeling the relationship between ingress motion and ingress discomfort ratings. The method presents a modeling framework that predicts subjective responses using human motion trajectories. The framework integrates curve alignment and data dimension reduction methods into the prediction model development. A case study is shown to demonstrate that human motion prediction models are more effective than simpler, more common ingress discomfort prediction models. The third part of this dissertation proposes a method for statistical hypothesis testing and sample size calculation for comparing ingress discomfort proportions of different vehicle designs. A dual-bootstrap method is proposed to estimate the standard deviation of ingress discomfort proportions estimated using a human motion prediction model. The proposed method is capable of separating the two sources of variation; the modeling variance, which results from the uncertainty in the estimated prediction models, and the sampling variance, which arises due to the randomness in the prediction dataset. The effectiveness of the proposed method is demonstrated through an ingress case study. The research presented in this dissertation is applicable beyond the analysis of ingress motion data; it can be applied to many fields where human motion data is available. At a broader level, the research presented can be useful in the analysis of functional data of many types, with particular applicability to multi-channel time-series data.