Monitoring and Command System with Network Video Camera

利用网络摄像机采集的视频流，在监控区域内发现运动目标，并将每个目标的位置、形状和时间戳传送给监控指挥系统。该系统为每个对象构造一个事件图表达式，该事件图表达式具有一系列带有时间戳的对象位置。表示对象的危险状态的事件图存储在系统中。系统确定输入事件图与存储的一些事件图匹配。该决定是根据停留时间、停留区域和对象的行为来执行的。我们提出并评估了一个三维位置测量使用一个单一的相机。根据我们的实验，如果监视区域是大约10m^2的平面，我们可以得到一个物体的三维位置，只有几个百分点的误差，只有一个摄像机。结合运动目标检测、事件图生成和报警系统，构建了停车场监控指挥系统的原型系统。当热鲁 关于我们 关于人和汽车的区域和行为的事件被预先登记在系统中。作为危险事件，指定了汽车的碰撞和人在停车场与汽车长时间停留。首先，从网络摄像机捕获的视频中检测运动对象。带有时间戳的对象的3D位置作为事件传输到系统。使用具有时间戳的3D位置，确定在指定区域中的碰撞和长时间停留。如果物体的状态与系统中登记的危险事件一致，系统发出警报或警告。如果人类在其中一个区域停留超过2分钟，系统会发出警报。提出了一种新的运动目标检测算法，并进行了测试。结果，通过物体的3D位置比通过2D位置能够更精确地执行碰撞检测，并且当发生指定事件时，正确地发出所有警报。少

Finding moving objects in a monitoring area using a video stream captured by a network video camera, the position, shape of each object with a time stamp are transferred to a monitoring and command system. The system constructs an event graph expression for each object with series of object positions with time stamp. Event graphs that represent dangerous states of objects are stored in the system. The system determines an input event graphs matches some event graphs stored. This decision is carried out based on the duration of stay, stay area, and behavior of objects. We proposed and evaluated a 3D position measuring using a single camera. According to our experiment, if the monitoring area is about 10m^2 a plane, we can obtain 3D position of an object with only a few percents error using just one camera. Integrating moving object detection, event graph generation, and alarm issuing systems, we construct a prototype system for monitoring and command system for a parking space. Dangerou … More s events with regard to areas and behaviors of humans and cars are registered in the system in advance. As dangerous events, collision of cars, and a long stay of human in a parking lot with a car are designated. At first moving object is detected from a video captured by a network camera. The 3D position of the object with time stamp is transferred to the system as an event. Using the 3D position with time stamp, the collision and long stay in the designated area are determined. If the state of the object is coincident with the dangerous event registered in the system, the system issues an alarm or warning.In the experiment, the 20 parking slots are registered as dangerous area. If human stays in one of the area more than 2 minutes, the system issues an alert. A new algorithm for detecting moving objects for long term monitoring is proposed and tested. As a result, collision detection can be more accurately performed by the 3D position of objects than by the 2D position and all alarms are correctly issued when the designated events are occurred. Less