Trajectory data travel mode identification method and device based on improved PointNet

Abstract

The invention discloses a trajectory data travel mode identification method and device based on an improved PointNet, and the method comprises the steps: carrying out the preprocessing of the original trajectory data of the travel of a traveler, and dividing a data set obtained after the preprocessing into a training set, a verification set and a test set; constructing an improved PointNet network model according to the size of the data dimension in the data set obtained after preprocessing; using data in the training set and the verification set to train the constructed improved PointNet network model, and obtaining a trained improved PointNet network model; testing the data in the test set by using a trained improved PointNet network model to obtain a corresponding travel mode identification result in the data in the test set; and correcting the obtained recognition result by adopting a preset post-processing algorithm.

Description

technical field [0001] This paper relates to the research and development of traveler transportation service technology, especially a method for identifying travel modes based on improved PointNet trajectory data. Background technique [0002] Travel mode identification of travelers is the key to the research and development of travel service technology, and is an important technology in the fields of traffic planning and traffic management. With the popularity and popularity of portable electronic devices (such as smart phones), the identification method of travel mode based on trajectory data has replaced the traditional method of household survey and telephone interview. Current research in this area is mainly based on a two-stage recognition framework, which first divides travel segments into travel-mode segments that contain only one travel mode, and then classifies the travel-mode segments segment by segment. However, there are problems with the current research metho...

AI Technical Summary

Problems solved by technology

However, there are problems with the current research method: First, the two stages of research are independent of each other. In practical applications, it is difficult to find the identification bottleneck when the two stages are implemented at the same time to identify unlabeled data, that is, if the recognition effect is not good, it is difficult to determine Is it because the division of travel segments in the first stage is inaccurate, or the classification model in the second stage is too sensitive to the division results?

Second, if there is an error in the division of travel segments in the first stage, it will affect the classification results of the second stage

Third, travel mode identification depends on the information of the front and rear track points, and the length of the track segment is required not to be too short. However, the two-stage method needs to divide the travel segment before classification, and the divided travel segment is too short to accurately predict the travel mode of the newly added track point, resulting in real-time Excessive bias in travel mode forecasts

Images