| Digital video surveillance system (DVRs), integrated with compute, multi-media, communication and network technologies, has played significant role in more and more fields. The encoder end of DVRs, with responsibility for video capture and encoding, is the key part of DVRs and straightly affect the performance and quality of the whole system. With the demand to achieve higher image quality , lower bit-rate and intelligence for automatically surveillance, digital video surveillance technology based on the MPEG-4 has become the mainstream of new generation DVRs.In this thesis, a hardware and software solution was proposed for implementation of 16 channel DVRs in the embedded X86 platform, in which, hardware took charge of the 16-chanel video capture and MPEG-4 encoding and encoded stream was transferred to PC via PCI interface. Software implemented the hardware device drivers, real-time demuxing of 16-channel system stream and real-time decoding of one channel of video streams. Besides, a simple but effective error concealment method was studied to ensure decoder to work well in the bit-error environmental. The running result shows that our hardware and software work well and stably.For the purpose of automatic alarm in the DVRs, a novel temporal-spatial change detection method, which was mainly depended on the motion vector information, was proposed for the detection of moving objects from the MPEG compressed field. Firstly, a temporally average filter and a spatially weighted vector median filter were applied to the motion vectors to alleviate the detection error due to the difference between real motion information and the motion vector. For Inter-MB, a temporally statistical modal with regard to vector angle was established and the moving object was detected by using the neyman-pearson criteria. For Intra-MB, a temporal compare was applied for distinguishing the motion MBs from uncovered background MBs. Experiment results show that the considerable detection could be expected. |
PDF Full Text Request