| I&T Solution |
Cloud Based Crowdedness Detection Solution
(REF: S-1518) |
| Trial Project |
|
| Solution Feature |
- A deep learning model for visualizing crowd density in high-stack images will be developed based on the method of multi-object detection
- A Transformers-based peak crowd prediction system uses the current moment, weather, and relevant public news to calculate a baseline for crowdedness
- The software platform is compatible with different suppliers' cameras based on the streaming media protocol, publishes the calculation results on the government webpage and provides UI interface
- The software runs on the cloud service platform of the Hong Kong Productivity Bureau, which runs on the Amazon cloud and provides 24/7 service
|
| Trial Application and Expected Outcome |
- Model: Develop a deep learning model that can estimate crowd density. The expected outcome is a trained model that can accurately estimate crowd density from images, as verified by a testing dataset. Success will be measured by the model's confusion matrix and F1-score on this testing dataset.
- Augmentation and Preprocessing: The expected outcome is an improved model that maintains high accuracy when tested on a diverse set of images. The success can be verified by comparing the model's performance on the original testing dataset and a new testing dataset that includes augmented and preprocessed images.
- Video Frame Extraction and Processing: Extend the system to handle video input by implementing video frame extraction. The expected outcome is a system that can estimate crowd density from video input. Verified by testing the system on a set of videos and comparing manual counts or ground truth.
- Scalable Pipeline Development: Develop a scalable pipeline that can handle video input from HD CCTV. The expected is a system that can process video feeds from multiple sources and store the crowd density estimates for further analysis. Verified by load testing the system can handle the expected volume of data.
- System UI Development: Develop UI for the system that allows users to view real-time crowd density estimates, alerts, and historical reports. The outcome is a UI provides a user experience and meets all the stated requirements. Verified by user testing to ensure the UI is intuitive and provides the necessary.
|
| Info on I&T Solution Provider |
| Solution Provider | : | TDU @ HKPC | | Address | : | HKPC Building, 78 Tat Chee Avenue, Kowloon, Hong Kong | | Contact Person | : | Dr. Cheng Jiefeng |
| Position | : | Senior AI researchor | | Tel | : | 95871037 | | Email | : |
geoffcheng@hkpc.org |
|
