Enhancing Public Safety: Fusing Skeletal Models and Generative AI for Robust Aggression Detection

In the pursuit of bolstering public safety through intelligent surveillance, Fujitsu's past endeavours leveraged skeletal models in conjunction with classical machine learning algorithms for Vision AI to detect aggressive and anti-social behaviour. While this approach demonstrated potential in recognising human interactions, it was marred by a high rate of false positives and challenges stemming from occlusion, variable lighting conditions, and other environmental factors. To address these limitations, Fujitsu has pioneered innovative platforms that integrates Generative AI with their existing skeletal model-based framework. This synergy has dramatically reduced false positives while significantly enhancing both precision and recall in aggression detection using newer approach on our codes and algorithms. This delves into the architectural intricacies of this fusion, elucidating how Generative AI complements skeletal models to overcome traditional Computer Vision barriers, thereby setting a new benchmark for reliable and scalable public safety applications and tools.