| I&T Solution |
Roadside Signboards Detection, Recognition, Identification an Structural Analysis
(REF: S-1746) |
| Trial Project |
|
| Solution Feature |
- We proposed a three-step approach for evaluating the potential signboard falling risks, which integrates cutting-edge AI techniques alongside physics-based modelling to enhance risk evaluation precision.
- The signboard segmentation module leverages Mask-RCNN and neural style transfer, which ensures accurate high-resolution segmentation within varying weather and lighting conditions.
- The defect detection module is based on Co-DERT and RT-DERT models, the best performing algorithm, which can efficiently pinpoint a range of defect.
- The risk assessment module is a fusion of AI and physical modelling: quantifying signboard deformations by comparing to past patrolling data, and further augmented by finite element analyses and Bayesian Inference.
- A user-friendly interface and backend system will be developed, facilitating easy signboards lookups and supporting data reflow.
|
| Trial Application and Expected Outcome |
- Create a curated signboard segmentation dataset with annotations to facilitate the training and assessment of the signboard segmentation model.
- Establish a labelled signboard defect dataset encompassing various defects like rust, deformities, missing displays, among others, to train and evaluate the defect detection model.
- Generate a dataset for signboard falling risks derived from outcomes of finite element modeling and Bayesian Inference, for training and evaluating the risk assessment model.
- The overall risk assessment accuracy will be evaluated using the data of historical signboard falling incidents happened in Hong Kong or other cities.
|
| Info on I&T Solution Provider |
| Solution Provider | : | The Hong Kong University of Science and Technology | | Address | : | Room 3564, 3/F, Academic Building, HKUST, Clear Water Bay, Hong Kong | | Contact Person | : | Walter Zhe Wang |
| Position | : | Assistant Professor | | Tel | : | 23588753 | | Fax | : | 23581534 | | Email | : |
cezhewang@ust.hk | | Webpage | : | https://walterzwang.github.io/ |
|
