Introduction:
During my second project in internship, I worked on an intriguing project focused on identifying unauthorized individuals entering the power rooms of HDBs (Housing Development Board) in Singapore under the supervision of Assoc Prof Yuen Chau. This project presented unique challenges as it required detecting unauthorized people amidst normal activities such as walking and working. By leveraging existing face recognition models and our proprietary algorithm, we aimed to develop a reliable and lightweight solution that could be deployed on edge devices. In this article, I will share the key aspects and accomplishments of this internship project.
Problem Background and Objectives:
Ensuring security within the power rooms of HDBs is crucial to prevent unauthorized access. The project aimed to identify and alert authorities if any individual entered the power room without authorization. The database of authorized personnel should be able to update at anytime with only one photo for a person, most the time, ID photo. We deployed CCTV cameras to monitor the area continuously. We divided the project into:
1. Face Detection and Filtering.
2. Face Recognition and Unauthorized Person Identification.
3. Real-time Upload and Storage.
Methodology:
Due to confidentiality agreements, I cannot disclose the specific codes and algorithms used. However, I can provide an overview of the general methodology we followed.
We began by developing an algorithm to detect and filter faces accurately from the video feed, addressing challenges such as face movement. Next, we utilized existing face recognition models and our proprietary algorithm to generate face embeddings—vector representations of facial features. We worked with the embeddings in the 128D space. By comparing these embeddings with those of authorized individuals in the database, we could determine if a person was authorized or unauthorized. Finally, we integrated a system for real-time upload and storage of the identification results, ensuring prompt action and analysis.
Results and Achievements:
Throughout the project, our team made significant progress in identifying unauthorized individuals entering the power rooms of HDBs. By combining face detection, face recognition models, and our proprietary algorithm, we achieved reliable results in real-time. The system effectively captured and filtered faces, recognized individuals using their face embeddings, and promptly uploaded and stored the identification results for further analysis and appropriate action.
Conclusion:
The internship project on unauthorized person identification within power rooms showcased the importance of advanced face recognition techniques and efficient real-time processing. Despite the limitations on disclosing specific codes and algorithms, our team successfully developed a robust system that could detect unauthorized individuals amidst normal activities. This project honed our skills in face detection and recognition, algorithm development, and real-time data management. The knowledge and experience gained during this internship project will undoubtedly shape our future endeavors in the field of computer vision and security systems.
Comments