PROJECT TITLE :

A Multi-Scale Attributes Attention Model for Transport Mode Identification

ABSTRACT:

Transport mode identification, also known as TMI, is extremely important for facilitating an understanding of urban mobility patterns and the choice behaviors of passengers with the end goal of improving urban transportation systems. TMI works by inferring the travel modes of user trajectories. Existing TMI methods typically rely on mobility features obtained from densely sampled GPS trajectory points (for example, one second per GPS point), as well as the data measurements of additional inertial measurement unit (IMU) sensors, in order to achieve a higher level of accuracy (e.g. accelerometer, gyroscope, rotation vector). However, this results in a significant increase in the amount of energy that is consumed by the mobile devices used by the users. In this paper, we propose a novel deep learning framework that we call the Multi-Scale Attributes Attention (MSAA) model. The purpose of this model is to extract discriminating trajectory features from GPS data alone, without the need to increase its sampling rate. The trajectories are partitioned into different scales as the first step of the proposed model, and then the model extracts the latent representation of local attributes at each scale. The MSAA model uses a Convolutional Neural Network (CNN) to capture the spatial correlation of different trajectory segments. It also uses an attention mechanism to select the most suitable local attributes on the various trajectory scales that can effectively characterize the various transport modes. These two components work together to create an accurate representation of the trajectory data. An ensemble model based on Neural Decision Forest (NDF) is used to fuse the heterogeneous features consisting of both measurable quantities and non-measurable elements for the purpose of determining the transport mode. This is necessary due to the fact that the learned latent local attributes are significantly distinct from the global features (for example, average, minimum, and maximum travel speeds, all of which are measurable quantities). Experiments on real-world datasets demonstrate the competitive performance of the proposed approach in comparison to several state-of-the-art baselines, with average improvements in accuracy ranging from 0.76% to 6.4%. This is demonstrated by the fact that the proposed approach achieves competitive performance. In addition, the multi-scale local attributes that were proposed are a good complement to the global characteristics. According to the findings of our research, the detection performance was enhanced by 2.3% on average when local attributes were incorporated into the analysis rather than using only global features.