Contest No. 1

Geospatial Information Systems

Title : "Spatio-Temporal Pattern Recognition on Trajectory dataset using Deep Learning" 

The Challenge Task

In our increasingly connected and sensor-rich world, trajectory data has emerged as a fundamental source of information for understanding mobility and behavior. It represents the digital footprint of moving objects over time, consisting of a sequence of spatio-temporal points with coordinates and timestamps. The Geolife GPS Trajectory Dataset, developed by Microsoft Research Asia, is one of the most famous publicly available human trajectory datasets. It provides unprecedented insights into real-world movement patterns, capturing a variety of transportation modes, and daily life activities, offering a holistic view of human mobility.

Trajectories are more than just paths on a map; they are rich, multi-dimensional data streams that encapsulate spatio-temporal behavioral patterns. Users exhibit diverse mobility patterns including regular commuting, occasional travel, recreational movements, and complex multi-modal transportation usage. The data also reflects periodic and context-dependent behaviors through different seasons, days of the week, and times of day. In addition, a subset of the data includes transportation mode labels, providing valuable ground truth for learning for behavioral pattern discovery. Recognition of these patterns is crucial for a wide range of applications, such as intelligent transportation systems, urban planning, and location-based services.

The DeepTrajectory Challenge 2025 of the 8^th ISPRS Geospatial Conference 2025, organized by the School of Surveying and Geospatial Engineering, University of Tehran, invites participants to apply deep learning algorithms for spatio-temporal behavioral pattern recognition using Geolife trajectories dataset.

Participants will be expected to:

• • preprocess raw GPS trajectories through noise filtering, outlier removal, and segmentation into meaningful movement episodes
  • extract advanced movement features such as velocity profiles, acceleration patterns, direction changes, and stop characteristics
  • transform the processed trajectories into formats suitable for deep learning architectures that preserve both spatial and temporal relationships
  • recognize and interpret spatio-temporal behavioral patterns and validate model outputs against real-world contexts

 

How to get the data:

The challenge uses a subset of the Geolife dataset focusing on users with over 100 individual trajectories. This selection maintains the richness and diversity of the collection while providing sufficient data volume for training sophisticated deep learning models. The dataset will be partitioned into training, validation, and test sets.

Download Link: https://www.microsoft.com/en-us/research/publication/geolife-gps-trajectory-dataset-user-guide/

 

Deliverables:

After producing the final result, the following materials should be provided in zip format:

• • technical report in docx or pdf describing all detail analysis (NOT more than 5000 words)
  • produced results
  • source codes used for preprocessing, modeling, and evaluation

 

Evaluation Protocol:

The DeepTrajectory Challenge 2025 invites researchers, students, and professionals to develop innovative deep learning solutions for spatio temporal behavioral pattern recognition in Golife dataset. Participants will be evaluated based on innovation and technical performance of the used deep learning algorithms on the dataset.