Advancements in Crime Detection: Bridging Domain Gaps with Dual Variational Generation

The objective of this project is to implement facial recognition technology on a dataset consisting of individuals from India in a heterogeneous environment. To address the absence of an Indian dataset, web scraping techniques are utilised to collect photos of players from a league’s official website. These players included both Indian and foreign nationals. Subsequently, a Dual Variational Generator is applied to the web-scraped photos to augment the training dataset. A framework of Haar Cascade and LBPR algorithm is utilised for face recognition. The combination of Haar cascades, LBPH algorithm, and DVG approach creates a robust framework for facial recognition in security applications. This framework offers a powerful blend of simplicity, efficiency, and accuracy that is specifically designed to address the issues faced in real-world deployments.