| Project Reference: | ITP/001/25LP |
| Project Title: | Improving Radar Detection Accuracy by Verification Method using Reinforcement Learning |
| Hosting Institution: | LSCM R&D Centre (LSCM) |
| Abstract: | Nowadays, despite the availability of numerous detection and monitoring sensing devices on the market, radar remains an effective tool. In two of our applied research projects, we utilize radar in the following ways: Traffic Control: Radar is employed to estimate the number of vehicles in a surveyed area. Security Monitoring: In a remote area, radar is utilized to detect movement and potential threats from animals or humans, ensuring safety while maintaining privacy. In general, radar offers the following advantages when used as a detection and monitoring tool: 1. Privacy Protection: Avoids capturing identifiable images, easing privacy worries in secure zones. 2. Swift Real-Time Response: Enables rapid decision-making in traffic control and security with immediate feedback. 3. Extensive Coverage: Monitors large areas up to 180 degrees and 3-4 kilometers from a single unit. 4. Cost Efficiency: Economical alternative to deploying multiple optical cameras, reducing overall expenses. However, in recent radar system applications, we encountered: 1. AFCD project issue: Radar mistaking sea waves for human figures, causing false alarms in marine park security. 2. Traffic Department project challenge: Radar confusing moving tree leaves with small vehicles, hindering traffic monitoring. To address the challenges faced by radar technology in accurately identifying objects, we propose a dual approach that utilizes clustering algorithms and reinforcement learning for verification. 1. Use DBSCAN and Isolation Forest for data clustering. 2. Implement reinforcement learning for verification using an optical camera agent. Ultimately, the optical camera is not essential for the production run. The proposed methodology aims to achieve: 1. Methodology for enhancing radar object detection. 2. A fine-tuned parameters specifically optimized for a radar applications. By integrating these advanced techniques, we aim to significantly enhance the reliability and accuracy of radar technology in both security and traffic monitoring applications. Further, we plan to publish our research in an academic journal. The tentative title of our work is "Improving Radar Detection Accuracy by Verification Method using Reinforcement Learning". |
| Project Coordinator: | Dr Chun Hung CHENG |
| Approved Funding Amount: | HK$ 2.5M |
| Project Period: | 6 Mar 2025 - 5 Mar 2026 |
