| Based on many technologies, such as computer vision and image processing technology, Video intelligent surveillance can not only capture and compress the real-time video, but also can detect and identify a certain person or object, then determine whether there is a abnormal thing, and finally send out alarms and provide useful information in the quickest and best way. It has brilliant market prospect. Taking into account some requirements such as the cost and size of the surveillance system, it is always implemented by embedded system. In this thesis, we design and realize the video intelligent surveillance on Ti Davinci platform at first. Then we focus on the research of transmission protect scheme which will be used to protect coded video.We first choose Ti Davinci dual core (ARM+DSP) platform to be the hardware implementation of video intelligent surveillance system, according to the needs of the system. After that, we compare the current algorithms for target detecting, target tracking and video compressing, and then choose the most suitable algorithms for the three mentioned procedures, based on the processing capability of the embedded system and the target of the system.After determining the hardware platform and algorithms, we design the system framework according to the data flow direction.Based on the analysis of the association and succession of tasks and the resources of Ti Davinci platform, we design the multi-thread software scheme. This scheme use multi-threaded technology, two levels of Ping-pong buffers and semaphores to make the dual cores run simultaneously. And this makes the system much more efficient. The scheme contains three threads: the front end processing thread, the video processing thread and the TS packaging and sending thread. In the front end processing thread, the onchip device named Resizer is controlled by ARM to transit video format. To realize this, we develop Resizer's driver. This thesis presents the API of this driver and core codes of this thread. For the video processing thread and the TS packaging and sending thread, we describe the way to realize them and present the definitions of key parameters and important codes. Then, the system is tested by standard clips and the results prove its real-time performance with CIF resolution video. We also verify the validity of the system by testing it in the pratical video analysis application scenarios.In remote video surveillance, the transmission delay of the network, bandwidth fluctuation, congestion, and channel fading often lead to packet loss. In order to reduce the image quality degradation brought by the packet loss, we research FMO and UEP. Based on old UEP model assisted by FMO[31], we added a variable FEC scheme and develop a new UEP model with FMO. At first, this model supposes there is an"Importance"for every MacroBlock. And then, based on the value of"Importance", it partitions the original picture into two Slices. After that, it protects these two Slices respectively based on different FEC. Finally, it lets the handled data go through the limited channel. We mathematically analyze this model and propose specific scheme to minimize the loss of"Importance"of the whole picture. Then, based on a particular concealment scheme, we propose a certain type of"Importance". At last, we simulate the scheme, and the results prove the feasibility of this model. |
PDF Full Text Request