Uncovering the Root Causes of Truck Rollover Crashes on Highway Ramps
Project Timeline: Jan 2021-Mar 2023
Links
Z. Bhuyan, Q. Chen, Y. Xie, Y. Cao and B. Liu, "Modeling the Risk of Truck Rollover Crashes on Highway Ramps Using Drone Video Data and Mask-RCNN," 2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC), Bilbao, Spain, 2023, pp. 4052-4058, doi: 10.1109/ITSC57777.2023.10421999. IEEE Xplore
Vehicle detection using modified Mask-RCNN: Oriented Bounding Box (OBB) + instance segmentation for analyzing truck rollover incidents on highway ramps.
 |
 |
| Andover, MA: Interchange of I-495 and I-93. | Auburn, MA: Massachusetts Turnpike Near Exit 10. |
Technical Approach
- Modified Mask-RCNN architecture
- Oriented Bounding Box (OBB) detection
- Instance segmentation techniques
- Drone video data analysis
- Highway ramp safety assessment
The deep-learning vehicle detection module plays a crucial role in identifying vehicles within the input data, while the tracking algorithm enhances the tracking precision by focusing on the orientation of vehicle bounding boxes. This integrated approach aims to provide a robust and effective solution for modeling the risk of truck rollover crashes on highway ramps using drone video data and Mask-RCNN.
Traditional tracking algorithms are typically tailored for horizontal bounding boxes, posing a challenge for scenarios requiring orientation-aware tracking, such as ours with Oriented Bounding Boxes (OBB). Our specific video data characteristics, lacking abrupt occlusions or random object disappearances, allowed us to design a minimal and efficient tracking algorithm tailored to our needs. This algorithm relies on the intersection-over-union ratio of a given vehicle in consecutive frames to ensure accurate and effective tracking.
