An automated video surveillance system using Viewpoint Feature Histogram and CUDA-enabled GPUs

This paper presents an automated video surveillance system which deals with content monitoring and activity change in the environment. We use Viewpoint Feature Histogram, an image descriptor for object recognition and pose estimation for the purpose of monitoring in the surveillance system. In order to enhance the performance of the system, we exploit the GPU architecture to perform data intensive task of surveillance system and implement it on CUDA-enabled devices. The experimental evaluation on the static data sets and live scenes captured from Microsoft Kinect show that Viewpoint Feature Histogram can be successfully used as an image descriptor in surveillance systems. We also test the performance of the Viewpoint Feature Histogram generation for different data sets on GPU and CPU to conclude that GPU clearly outperforms CPU for larger datasets.