| Video surveillance is defined as security surveillance and remote-control area, the use of video capture and transmission technology to monitor real-time monitoring of those regional images, sound and other features of the signal of a monitoring technology. Video surveillance technology has mainly experienced three development stages are: stage of analog video monitoring technology, digital video surveillance technology, stage, stage of network video monitoring technology.Widely used in TV picture tubes as early as the period, it produced a simulation of video surveillance technology, which uses the camera and monitor one-connection structure, using the video cable transmit analog signals as the transmission media, which is closed-circuit monitoring system initial stage. Because of this structure is a one to one waste of resources and another that followed a simple hardware circuit control implementation of video switcher, which can use a monitor to connect multiple video cameras, has since introduced a zoom lens, and PTZ. Because computers can only handle digital signals, it needs to be collected into an analog signal is converted to digital signals. The amount of the advantages of using digital signals are: ease of digital signal compression and storage, video quality is higher, which is not external electromagnetic interference, suitable for long-distance transmission, a binary digital signal to facilitate queries. Digital signal networking is relatively simple and easy to expand. At the same time using computer programming as monitoring and management system, monitoring and management system with greater portability and reliability. Through the network will monitor an unlimited expansion of the scope to achieve metro monitoring, wide area monitoring. The same time, the video stored on the remote server, even though the local man-made device has been forcibly destroyed, remote video storage information will be strong evidence can not be destroyed. The current field of video surveillance to the network is in the localization of the transition phase.CRC "Intelligent Control Eyes" is an Internet-based professional network video surveillance operations, which is CRC network video surveillance business brand. It is based on the broadband network to provide customers with images, sounds, and various alarm signals the remote collection, transmission, storage, processing and transmission of a new value-added telecom services. Through this service, users can free the time and place restrictions, easy targets for monitoring real-time monitoring, supervision and management and video store, you can also monitor the site layout of alarm equipment and monitoring devices to achieve alarm linkage. Users through web pages (client browser platform) can be targets for monitoring real-time monitoring, security management, and environmental watch such an operation. Although the network address of the remote video surveillance monitoring and real-time storage problem, but with the use of diversification, the user is often not satisfied with a simple remote PTZ control, focal length scaling operation, the original video surveillance function of the system stretched. Network video surveillance in certain sectors of the special needs of the business has not been effectively resolved, the traditional network video surveillance is often not achieve the following common features: no duty function, selective storage capabilities, real-time alarm. If the Network Monitor in to add motion detection function, these problems will go away. Therefore, this article aims at present a secondary development of solutions, in the CRC "Intelligent Control Eyes" by adding motion detection module.Intelligent motion detection is a core part of the monitoring system, which is how quickly and accurately detect surveillance video of the moving object. Traditional motion detection There are three main methods: optical flow method, the adjacent frame difference and background subtraction method. The basic idea of optical flow method are: the optical flow is the space, moving objects in the observation plane pixel instantaneous velocity, given all of the pixels in the image is given a velocity vector, the plane through the projection of each image pixel and three-dimensional space motion picture counterpart, which form a pixel of the playground, called optical flow field. When the scene contains moving objects, when some of its instantaneous velocity and the velocity vector of the adjacent background is different to detect a moving object the size and location. However, the objective factors, such as multi-light source, shadows, transparency, occlusion and noise, making the calculated optical flow distribution is not accurate enough. Coupled with the calculation of optical flow algorithm is more complex, time-consuming and more robust hardware and the poor have higher requirements. Optical flow method used alone is difficult to achieve real-time detection. Adjacent frame difference method The basic idea is: check the adjacent pixel intensity between the two changes, when the pixel intensity changes in the larger context, is often considered as caused by the movement of objects, according to the pixel intensity change discriminant movement of objects in the scene. Image acquisition and transmission process, the noise generative inevitable, the threshold is too small will set the scene noise, pixel mistaken for a moving object, the threshold set is too large will occur undetected. Adjacent frame difference in motion detection method has a relatively wide range of applications. Background subtraction prior real-time storage or a default background image for the background image sequence modeling, get background model. When the video capture to a particular frame, after subtracting the background model to calculate the value of a certain threshold within the limits of the current frame image and the background image to deviate from the larger pixels, larger than the threshold value of these deviate from the provisions of the pixels moving object . Non-adaptive background subtraction method, which need to address the background model and updating problems.Based on the above advantages and disadvantages of three methods, this paper proposes a motion-based method of historical images of the dynamic monitoring. The basic idea of this approach are: within the time prescribed in the recorded image trajectories of moving objects. Movement of the pixels is set to the current timestamp, while the exercise took place over the long-pixel was abandoned. Therefore, historical images of the movement describes the moving object within a specified time of observation trajectory. The level of entire paper is structured as follows: Chapter one first analyzes the video surveillance technology, the development of three phases: Video analog technology, digital video technology and network monitoring technology, and China Railcom "network accused of eyes" business made a brief introduction. Chapter two on the background knowledge to achieve development, such as network topology, protocol family, video, standards and norms, and networks such as the basic principle of the camera made a detailed introduction. Chapter three consists mainly of system analysis, through the establishment of UML modeling The implementation of the dynamic recognition module diagram, use case diagrams and activity diagrams, and the establishment of a "network accused of eyes" of the network topology. An outline of the development tools and development systems for further pave the way for the detailed design and coding. Chapter four of the dynamic recognition module of the detailed design, focused on analyzing the dynamic recognition of the traditional three methods: the frame difference method, optical flow the background subtraction method and advantages and disadvantages of three methods proposed based on Motion History Images method. MHI method of analysis of the principles and work processes. The fifth chapter is a procedural coding and testing part of the UDP-based protocol to do a faster network connection, through the white box and black box testing and found that the development process a number of threshold settings are not suitable for dynamic object recognition led to the problem of low precision, through the adjustment of the threshold to achieve the desired results. Finally, the entire article are summarized and prospects.With the continuous development of network technology, Internet-based monitoring would be the ultimate place of local monitoring techniques, and as the gradual popularization of 3G networks, the next need to give further consideration to cross-platform system migration and so on, enables customers in the right the process of monitoring the region to monitor not only the connection into the computer terminal on the Internet should be able to phone, PDA and other mobile devices to operate, and the resulting benefits are considerable. The same time, dynamic recognition can be sublimated, to join such as the AdaBoost algorithm face recognition technology, allows the system can automatically identify the monitor area, dynamic objects that are pedestrians, those non-human moving objects. Also can join the digital image recognition, or charges for car parks Exit automatic license plate recognition, and so more convenient features, making a systematic analysis of the results closer to human vision and way of thinking.Although the MHI-based approach can achieve better moving object in the video segmentation and gesture recognition, but because the duration of the selection of sports history, would make the selected split the objectives of a long tail phenomenon, can lead to division delay. At the same time, when the target region in the segmentation goal is too large or too much, also have multi-target adhesion problems. Can tell the program by testing a number of threshold settings on the test results plays a crucial role, so you can more accurately through the practice of trying to obtain the threshold value range, or simply part of the threshold value is converted by the global constant variables, defined by the user on their own values. Further development of the system self-learning function, making the threshold can be dynamically identified by the system through various use cases of the training myself to be the best value. As the dissertation is finished to time constraints, some of these problems still need to study and work in the future further resolved. |
PDF Full Text Request