CAJ | 학술논문

为了克服传统相邻帧差法和混合高斯模型的不足，提出一种结合改进的相邻帧差法和混合高斯模型的运动检测方法。对于相邻帧差法，增加了梯度比较和中值滤波，突出了运动目标边界和抗噪声干扰能力。通过结合相邻帧差法和混合高斯模型的结果，把图像分为4种区域，对每个区域内的混合高斯模型参数采用不同的更新策略，减少了运动目标对背景模型的干扰，对于物体在运动与静止之间转换的情况也具有很好的适应性。根据区域的划分情况和相邻帧差法的结果确定最终的运动区域。实验表明，既缓解了相邻帧差法产生的空洞问题，也消除背景物体突然转为运动物体后产生的“影子”问题。
위료극복전통상린정차법화혼합고사모형적불족，제출일충결합개진적상린정차법화혼합고사모형적운동검측방법。대우상린정차법，증가료제도비교화중치려파，돌출료운동목표변계화항조성간우능력。통과결합상린정차법화혼합고사모형적결과，파도상분위4충구역，대매개구역내적혼합고사모형삼수채용불동적경신책략，감소료운동목표대배경모형적간우，대우물체재운동여정지지간전환적정황야구유흔호적괄응성。근거구역적화분정황화상린정차법적결과학정최종적운동구역。실험표명，기완해료상린정차법산생적공동문제，야소제배경물체돌연전위운동물체후산생적“영자”문제。
To overcome the drawbacks of the traditional adjacent frame difference method and Gaussian mixture model,we present a mov-ing object detection method which combines the improved adjacent frame difference and Gaussian mixture model.For the adjacent frame difference it adds gradient comparison and median filter,stands out the moving object boundary and enhances the ability of noise interference resistance.By combining the results of adjacent frame difference and Gaussian mixture model,we divide image into 4 regions.Different up-date strategies are adopted to the parameters of Gaussian mixture models in each region,this reduces the interference of moving object on background model and has a better adaptability on the situation of objects switching between motion and stillness.The moving region is even-tually determined according to the partition situation of regions and the result of adjacent frame.Experiment demonstrates that the combination of two methods mitigates the “hole”problem of the adjacent frame difference,and also eliminates the “shadow”problem when a background object suddenly switches to moving object.