6G-Enabled Edge Computing for Real-Time Pedestrian Intent Analysis
| Reference number | |
| Coordinator | RISE Research Institutes of Sweden AB - RISE AB - Digitala System |
| Funding from Vinnova | SEK 99 946 |
| Project duration | February 2026 - September 2026 |
| Status | Completed |
| Venture | 6G - Competence supply |
| Call | 6G - Supervision of degree work |
Important results from the project
Yes, the goals were met. We successfully developed and evaluated models to predict pedestrian intent to enter a door. The analysis revealed that four simple trajectory features (distance, velocity, angle) account for almost all predictive signal, while body pose provided negligible improvement. A compact GRU model was trained, achieving ~87% balanced accuracy. Deployed locally on an NVIDIA Jetson Orin Nano, the entire pipeline comfortably met the 30 FPS real-time target.
Expected long term effects
The project enables smart doors that open only when a person intends to enter, reducing building energy losses and mechanical wear. The system validates the vision of 6G Edge Computing: by running inference on distributed, local edge infrastructure, it achieves the ultra-low latency required for real-time actuation. This also preserves personal privacy, as no video is transmitted to external clouds. This makes the technology commercially scalable and aligned with sustainability goals.
Approach and implementation
The project was conducted as a master´s thesis by Jesper Malmgren at RISE in collaboration with ASSA ABLOY. Visual features (trajectory and body pose) were extracted from ceiling-mounted door camera footage. These were systematically evaluated using SHAP analysis. Neural networks (like GRU and 1D-CNN) were then trained and ablated to find the optimal model. Finally, the complete system´s performance and latency were benchmarked locally on an NVIDIA Jetson Orin Nano edge device.