Police and border guard authorities often struggle to detect human trafficking victims who may appear outwardly “fine” or have been coerced into hiding their distress. Existing identification procedures typically rely on verbal interviews, straightforward document checks, along with cross-referencing with additional external intelligence when available, leaving a critical gap when victims are too fearful to speak or physically altered to mask their identities. Recognising this, the VANGUARD project is exploring new ways to spot concealed distress or disguised appearances, crucial capabilities that existing tools and procedures may not address adequately.

To this end, our research focuses on two complementary approaches to emotion recognition: facial expressions and body posture. By using deep learning models designed to flag subtle cues of fear, anger, or distress, we aim to augment police and border guard authorities’ ability to see beyond an individual’s words. This can prove valuable in situations such as interviews, conducted by experienced personnel with potential victims of trafficking, or at other similar lines of control. In parallel, the project studies face recognition methods robust enough to handle partial occlusions, swelling, or other injuries, a grim reality in trafficking situations. As a result, the manual task of identifying whether an individual appears on existing databases of known victims becomes automated, offering valuable time for the relevant authorities to act before it is too late. Overall, this research aims to provide officers with a prompt indication of whether someone in the crowd may be a known victim or whether a person of interest (e.g., potential victim) is exhibiting non-verbal signs of distress.

Throughout this process, ethical considerations guide both data collection and model development, ensuring that potential biases are monitored and addressed from the early stages. These lines of inquiry are still in the prototyping stage, with initial tests planned in controlled environments. The goal is to develop and refine proof-of-concept systems that can be tested in real-world conditions, without replacing or undermining existing safeguards around privacy and due process. By confronting the challenge of silent distress and hidden physical alterations, we hope to equip police and border guard authorities with meaningful tools that help prevent tragedy, uphold human rights, and make critical interventions before it is too late.