CAJ | 학술논문

基于GMM（Gaussian Mixture Model ，混合高斯模型）的物体识别基础上，利用GLCM （Gray‐level co‐occurrence matrix ，灰度共生矩阵）及基于GLCM 提取的纹理特征来解决当前运动目标检测所存在的问题，如动态场景的变化，光照突变及天气变化等。GLCM 在局部区域的往复运动具有相对不变性，因此利用这个特点对基于GM M 检测的前景进行再判断以解决动态场景的问题，将检测窗口中当前帧和前两帧的GLCM 特征值进行比较，如果其GLCM 特征值的差值小于给定的阈值，那么可以判断当前区域为背景，反之则为前景。图像的纹理特征具有抗光照突变性，经过分析其中的4个特征值并阈值化最终得到更加纯净的前景和更加准确的检测结果。通过CPU／GPU（Central Process Unit／Graphic Processing Unit）协同并行计算大大加速了运动目标检测过程。实验证明这种新的检测算法在检测精度和处理速度上比其他算法有明显改善。
기우GMM（Gaussian Mixture Model ，혼합고사모형）적물체식별기출상，이용GLCM （Gray‐level co‐occurrence matrix ，회도공생구진）급기우GLCM 제취적문리특정래해결당전운동목표검측소존재적문제，여동태장경적변화，광조돌변급천기변화등。GLCM 재국부구역적왕복운동구유상대불변성，인차이용저개특점대기우GM M 검측적전경진행재판단이해결동태장경적문제，장검측창구중당전정화전량정적GLCM 특정치진행비교，여과기GLCM 특정치적차치소우급정적역치，나요가이판단당전구역위배경，반지칙위전경。도상적문리특정구유항광조돌변성，경과분석기중적4개특정치병역치화최종득도경가순정적전경화경가준학적검측결과。통과CPU／GPU（Central Process Unit／Graphic Processing Unit）협동병행계산대대가속료운동목표검측과정。실험증명저충신적검측산법재검측정도화처리속도상비기타산법유명현개선。
For current moving object detection algorithms ,therexist three main issues ,inclu‐ding dynamic scene ,sudden illumination changes ,and bad weathers .These problems reduce the performance of video segmentation .The Gaussian Mixture Model algorithm is widely used cur‐rently .T he improved moving object detection algorithm based on GLCM for video surveillance is presented in this paper .The proposed algorithm computes the Gray‐level co‐occurrence matrix (GLCM ) to eliminate the effect of dynamic background by threshold firstly .Secondly ,it extracts the texture which has the property of resisting sudden change in light according to the GLCM .Fi‐nally ,cleaner foreground and more accurate results are obtained .What’s more ,the algorithm runs efficiently through CPU/GPU collaborative computing in parallel .