一种基于小波变换的运动估计方法

提出了一种基于运动边缘检测与连续小波变换相结合的运动目标检测与跟踪方法。通过多尺度运动目标边缘检测．提取出运动目标图像能量集中的边缘，得到抗噪性好、边缘连续清晰的运动目标。利用连续小波变换把目标边缘图像序列映射到运动状态参数空间进行分析。实验表明，该算法是对有噪、旋转和遮挡等复杂运动目标进行运动估计的有效方法。

Motion Estimation Based on Wavelet Transform

Strong noise and temporary occlusions make existing methods (such as optical flow, block matching and CWT methods) unable to make satisfactory detection and tracing of moving objects. To deal with strong noise and temporary occlusions, we propose improving CWT (continuous wavelet transform) method by combining it with multi-scale edge detection. We employ multi-scale edge detection to focus the image energy on the image edges, thus obtaining moving objects with clear and continuous edges and consequently of less noise. After that, we use CWT to map the input signals of edges of moving objects onto the motion-state-parameter space. Finally, we use three energy densities--speed, displacement, and scale--to estimate motion parameters. Our main contribution is the use of multi-scale edge detection before using CWT. Our simulation experiments show that using CWT algorithm alone cannot find the energy peaks of moving objects when noise is strong. Our simulation experiments also show that our improvement on CWT algorithm can make such energy peaks converge very quickly.