The PolyU Centre for Advances in Reliability and Safety (CAiRS) has created a state-of-the-art (SOTA) modified residual network with a colour attention module, trained on a customised ‘Blurveillance’ dataset. The small model size outperforms other SOTA networks in edge applications. It classifies images into five categories Normal, Natural Blur, Defocus Blur, Dirt Blur and Spray Paint Blur.
Smart surveillance cameras that are used in unconstrained environments can be tampered with due to either environmental factors or malicious human activity. This can blur the video content and make it difficult to identify what is happening in the scene. This can lead to mistakes and wrong real-time decisions.
With over 80 years of proud tradition and ranking among the world’s top 100 institutions, The Hong Kong Polytechnic University (PolyU) aspires to be a leading university with world-class research and education.
PolyU is a home for educating thinkers, discoverers, innovators and communicators in delivering positive impact. We are committed to nurturing tomorrow’s leaders today, through a holistic education that provides graduates unrivaled placements to thrive in communities, industries and businesses.