CAJ | 학술논문

为解决车辆阴影检测中易将车辆阴影相似的车辆区域误检测为车辆阴影的问题，提出了一种基于超像素和支持向量机的车辆阴影检测算法。首先，利用简单线性迭代聚类法将图像分割为若干超像素；然后，以超像素为基本检测单位，根据HSV空间中的一组判别条件对车辆阴影进行初步检测；在此基础上，利用支持向量机识别并去除被误检测为车辆阴影的车辆区域，进而得到最终的车辆阴影。实验结果表明，所提算法能够较好地区分车辆阴影及与车辆阴影相似的车辆区域，提高车辆阴影的检测率和分类率。
위해결차량음영검측중역장차량음영상사적차량구역오검측위차량음영적문제，제출료일충기우초상소화지지향량궤적차량음영검측산법。수선，이용간단선성질대취류법장도상분할위약간초상소；연후，이초상소위기본검측단위，근거HSV공간중적일조판별조건대차량음영진행초보검측；재차기출상，이용지지향량궤식별병거제피오검측위차량음영적차량구역，진이득도최종적차량음영。실험결과표명，소제산법능구교호지구분차량음영급여차량음영상사적차량구역，제고차량음영적검측솔화분류솔。
To solve the problem that the vehicle region similar to the shadow is apt to be wrongly de-tected as the shadow during vehicle shadow detection, a vehicle shadow detection algorithm based on superpixel and SVM( support vector machine) is proposed.First, the current image is segmented to several superpixels by the simple linear iterative clustering method.Then, according to a group of criterion conditions in HSV( hue, saturation, value) space, the vehicle shadow is detected prelimi-nary by taking superpixels as basic testing units.Finally, the vehicle region which is detected wrong-ly as the vehicle shadow is recognized by SVM and removed from the preliminary detection results, thus the final vehicle shadow is obtained.The experimental results show that the proposed method can well distinguish the shadow from the vehicle region similar to the shadow, and can improve the detection rate and the discrimination rate of the vehicle shadow.