An Online Face Clustering Algorithm for Face Monitoring and Retrieval in Real-Time Videos
Ye Cai, HaiYang Gan · 2019
Facial recognition in real-time videos will output a feature vector for each face tracking process, namely Euclidean embedding. The task of face retrieval is to find similar faces from the historical face database. Clustering historical face Euclidean embedding can improve retrieval performance. The data collected by the facial recognition system from real-time video is not all available. Since the traditional clustering algorithms operate under the condition that all face embedding data is available, it is categorized as an offline clustering method and offline clustering methods are not suitable for real-time scenarios. Based on real-time video surveillance scenarios, we propose an online clustering algorithm which can cluster the face embedding and achieve an accuracy of 84% on the benchmark video dataset. Our algorithm achieves better results than state-of-the-art online methods, and it is also comparable and even superior to the offline clustering algorithms.