Design And Realization For Embedded Network Camera Based On Content Retrieval

With the continuous development of camera technology and network technology, network camera has been widely used in various fields. It not only can be used for regional monitoring based on computer local network (for example, traditional monitoring such as the monitoring of community, office, bank, market), but also can be used for new type of cross-regional remote monitoring and on-line display through the Internet such as children and elderly care, unmanned engine room monitoring, tourist attractions online display, product exhibition, etc. And with the improvement of DSP technology, as well as the expansive application of embedded system, network camera performance has increased substantially. It becomes possible to the uptake images for further process.This paper starts from the specific application of image surveillance. It gives a design and implementation of network camera system based on content retrieval. This paper takes the process of embedded system development and design as the clues to the writing sequence, and introduces the design and realization of the network camera based on content retrieval.This paper proposed the development of the network camera and the application fields first, then, made out the function requirements of the system, and designed the functional modules focus on these requirements. These functional modules includes video sample module, image processes module, LCD display module, keyboard, memory module and net transport module. After the functional modules design, this paper chose the DSP as the system framework, then, focus on the weak expand ability, added CPLD to the system as the logic control. These logic control parts include the reset and read/write of peripheral modules, address decoder of asynchronous space and the key scanning algorithm. This makes the whole framework have a flexible expansion. And the compute ability of the CPLD can release the processor burden. After the decision of the system framework selection, this paper completed the expansion of the peripheral modules, and finished the hardware overall design for the system.With the hardware foundation, this paper, to the video stream as the perspective, introduced the every step of the process to the video frame and the data format before/after the process. It connected the modules of the system through once process to the video stream, and made these a whole, also gave the function of each module. For the process of the video stream, this paper gave an overall structure to the software algorithm, the system software flow includes three parts: sample of the video information, image pre-process and detection of moving object. According to the flow, this paper assigned the mission for the two core of the processor. The work for the core A is: drive the modules after power up, build the background model, pre-process the images (mainly about gray of image) and update the background model according to the new image frame. The work for the core B is: using the moving object detecting algorithm to difference the gray images and get the moving object part in the surveillance scenes, transport the interested images to the host through the Internet.After that, this paper focused on the design and implement of hardware platform, including each module's design in detail. First this paper introduced the DSP minimum system, and then gave the expansion for the other peripheral modules. The DSP minimum system includes clock, power and reset signal. The peripheral modules part mainly about memory device (including SDRAM, FLASH and SD card), video sample device, LCD interface and network device. The hardware platform part is mainly about how the chip pin connected. After that, this paper discussed the part of CPLD logic control including global address decoder logic, network logic, global reset logic, key scanning algorithm and the assignment of the asynchronous space address.After the hardware platform design, this paper designed and implement the drives for the devices. It made out the driver for the system core. The main function of this part is to give the support to the whole system, make sure that each module can communicate to the system core. These drivers include the configuration to the system work clock, the configuration to the EBIU, PPI and TIMERS. Then implement the drivers to the memory device, video decoder, LCD and network device.The software mainly includes three parts: sample of the video information and image pre-process, detection of moving object and transplant and optimize to the algorithm. The first part described the image format based on the video decoder. In the meanwhile, in order to display in the LCD, this paper described the conversion from YUV space to the RGB space. Then optimize the algorithm to run in the platform efficiently. Then it gave the algorithm of the conversion from the YUV space to the gray space. The second part discussed two moving object detecting algorithm: frame difference and background difference. Through the comparison between the two algorithms, this paper chose background difference as the system detecting algorithm. Then it gave the procedure of building the background image and updating it. This paper used statistic-based way to build the background image and updated it dynamically in the time of image sample. After building the background image, the moving object can be detected in the video frames using the difference and to get ready for the next work. The system took the binarization process to the gray image after difference so that the moving object in the image frames was easier to distinct. Combined the procedure of the image processing, this paper assigned the memory space for the images that were processed in each phase and this was throughout the whole image processing phase. The algorithm optimization part described the further optimization to the code in the ASM level combined the processor resources.In the last part, this paper summarized the entire system design and implement, and then it discussed the advantages and disadvantages of the system and put forward the main content to the next phase of work. In contrast, the advantage of this system is to transplant the image processing part to the nodes so that to reduce the host handling stress. In the meanwhile, it also could save the network bandwidth. There are, so far, some shorts in this system. These shorts include: the system resources are not fully used yet and the classifying and recognizing algorithms of the object are not transplanted to the embedded system yet. It is still processed by the host. In the next step, the main work will be around to improve system performance, saving network bandwidth and other content.