【摘要】 针对当前三维场景下监控视频与空间信息融合动态性差、精度低的问题，基于直接线性变换与图像动态匹配原理，提出一种监控视频与三维空间信息动态融合的方法。通过设置预设帧构建监控视频初始图像与三维空间的映射关系，并依据视频图像平移、俯仰和缩放的值建立图像索引；在实时监控时，采用加速鲁棒特征算法进行图像匹配，获取控制点；针对监控图像变化时三维空间映射控制点个数不足的情况，提出了大、小尺度场景映射策略；采用直接线性变换算法，实现了三维空间信息与监控视频的动态融合。为了验证方法的正确性，以海康威视球形摄像机为采集设备，选取某高校平坦地面和坡地的区域进行实验验证与方法对比。结果表明，该方法在相机任意平移、俯仰和缩放的状态下能够实现动态精准的融合三维空间信息，误差在13个像素以内，满足实际应用场景。更多还原

【Abstract】 To address the issues of poor dynamism and low accuracy in integrating surveillance video with spatial information in 3D scenes, this paper proposes a dynamic fusion method based on Direct Linear Transformation(DLT) and image dynamic matching principles. First, preset frames are established to construct the initial mapping between surveillance video images and 3D space, and an image index is built based on the pan, tilt, and zoom values of the video. During real-time monitoring, the Speeded Up Robust Features(SURF) algorithm is employed for image matching to obtain control points. To address the problem of insufficient control points for 3D space mapping when surveillance images change, a large-and small-scale scene mapping strategy is proposed. The Direct Linear Transformation algorithm is then used to achieve dynamic fusion of 3D spatial information with surveillance video. To verify the correctness of the proposed method, data collected by Hikvision PTZ cameras in a university area containing flat ground and slopes were used for validation and method comparison. Experimental results indicate that the proposed method enables accurate dynamic fusion of 3D spatial information under arbitrary camera pan, tilt, and zoom conditions, with an error within 13 pixels, meeting the requirements of practical applications.更多还原