A Method for Pedestrian Trajectory Prediction Using INS-GNSS Wearable Devices

Area of Science:

• Robotics and Artificial Intelligence
• Geospatial Navigation and Tracking

Background:

• Pedestrian trajectory prediction faces challenges due to spatiotemporal uncertainty, limiting the accuracy of current machine learning models.
• Advancements in AI are driving a shift towards neural network-based autonomous decision-making frameworks for trajectory prediction.

Purpose of the Study:

• To develop a robust multi-source perception fusion system for accurate pedestrian trajectory prediction.
• To enhance pedestrian localization and future movement forecasting by integrating inertial measurement units (IMUs) and global navigation satellite systems (GNSS).

Main Methods:

• Proposed a Gait Adaptive UKF (Gait-AUKF) for improved localization by fusing IMU and GNSS data, adapting to pedestrian gait patterns.
• Developed a multi-source fusion attention mechanism framework using GRU and LSTM for trajectory prediction, incorporating an A* path planning algorithm.
• Utilized wearable INS-GNSS devices for high-precision data acquisition.

Main Results:

• The Gait-AUKF significantly reduced localization errors: eastward (30%), northward (26.27%), and vertical (49.08%) compared to UKF and AKF.
• The complete prediction framework achieved substantial reductions in prediction errors: Average Position Error (APE) by 68.54% and Direction Error (DE) by 70.42% versus LSTM and Transformer models.
• Ablation studies confirmed that Gait-AUKF and A* path planning improved model performance, decreasing Average Displacement Error (ADE) by 68.49% and Final Displacement Error (FDE) by 71.86%.

Conclusions:

• The proposed INS-GNSS wearable system and fusion framework offer a significant advancement in pedestrian trajectory prediction accuracy.
• The Gait-AUKF and attention-based prediction model effectively address spatiotemporal uncertainties in pedestrian movement.
• This approach holds promise for applications requiring precise pedestrian tracking and forecasting, such as autonomous driving and intelligent surveillance.