Abstract
Surveillance cameras typically use a 2D vector graphics processor (VGPU) or a low-end 3D graphics processing unit (G PU) for user interface (UI) rendering. Since the 3D GPU supports a diverse array of programming languages, it can accelerate not only UI rendering but also computer vision algorithms. This paper presents a surveillance camera application that can effectively utilize low-end mobile GPUs. The proposed approach encompasses two key techniques: 1) coordinated pro-cessing to leverage the GPU operators, and 2) a multiple render target technique for multi- UI rendering to minimize memory bandwidth. The proposed approach increased a performance of 68% and reduced 16% of the memory bandwidth compared to the conventional approach.