局域生成对抗网络下视频动态监控系统设计

Design of video dynamic surveillance system under local generative adversarial network

针对当前使用的基于机器视觉、运动矢量空间编码的监控系统受到噪声影响,导致目标跟踪效果不佳的问题,提出了局域生成对抗网络下视频动态监控系统设计。通过嵌入式微处理器将任务集成在芯片内部,使设计趋于小型化。使用RS232型号作为云台控制器设备接口,对摄像头进行变焦或者拉伸等操作,使用流媒体服务器结构以流式协议直播给客户端。使用服务器中专门的线程对等待队列中的视频进行分配,重建动态视频超分辨率,计算对抗损失,通过损失函数的梯度交替训练及时更新参数,避免噪声影响监控效果,使视频监控内容与实际视频视觉上一致。由实验结果可知,该系统能够清楚拍摄到动态目标,且目标跟踪成功率最高为99.9%,具有良好的跟踪效果。

Aiming at the problem that the current monitoring system based on machine vision and motion vector space coding is affected by noise,resulting in poor target tracking effect,a design of video dynamic monitoring system under local generation countermeasure network is proposed.Through the embedded microprocessor,the task is integrated into the chip,which makes the design tend to miniaturization.RS232 model is used as the equipment interface of PTZ controller to zoom or stretch the camera.Through the streaming media server structure,live broadcast to the client with streaming protocol.The special thread in the server is used to allocate the video in the waiting queue,rebuilding dynamic video superresolution,calculate the countermeasure loss,and update the parameters in time through the gradient alternating training of the loss function,so as to avoid the noise affecting the monitoring effect and make the video monitoring content visually consistent with the actual video.The experimental results show that the system can clearly capture the dynamic target,and the target tracking success rate is 99.9%,which has a good tracking effect.