An object detection and tracking technique can be applicable for video contents analysis in such applications as intelligent surveillance systems and interactive broadcasting services. Most of methods utilize only raw pixel data for accurate performance (so called the 'pixel domain approach'), but they require customized hardware due to high computational complexity. On the other hand, the 'compressed domain approach', which use encoded information such as motion vectors, shows extremely fast computation time, but might have worse performance in complex scenes compared to the pixel domain approach. To overcome these limitations, my team in KAIST proposed both the dissimilarity minimization (DM) and the probabilistic spatiotemporal macroblock filtering (PSMF) algorithms as a hybrid algorithm of compressed domain and pixel domain for H.264/AVC compressed videos (SPIE 2009, LNCS 2007; view presentation). Since we make use of both partially decoded data and encoded information, the methods show reliable performance in natural scenes as well as fast computation time enough to be performed in real-time.
Dissimilarity minimization
probabilistic spatiotemporal macroblock filtering
Using deep learning and cloud technologies, we are actively seeking to develop an advanced AI engine for real-time scene recognition and understanding from streaming videos, which can be applied to a variety of IT industries, including surveillance system, digital media broadcasting, online shopping malls, and mobile apps.