Low Cost Visual Event Detection over JPEG Flows

For video surveillance systems, visual event detection is an interesting resource since it can reduce the volume of data to be transmitted or stored. Furthermore, visual event detection offers information about physical events occurring inside the scene. This article presents a low cost algorithm for visual event detection. The proposed algorithm is based on the partial decoding of the JPEG (Joint Picture Expert Group) flow produced by the IP cameras hardware and the analysis of the DCT (Discrete Cosine Transform) coefficients on contiguous frames. Two are the main advantages of this algorithm: It runs over a reduced memory space and it is inexpensive in terms of CPU resources. In order to show the advantage of the proposed algorithm, a visual event detection system has been implemented in C language over Linux.