LOBSTER与MOG结合的目标检测方法

An Object Detection Method Combining LOBSTER with MOG

借助深度相机数据一定程度上解决了目标检测中的颜色伪装问题,但又导致深度伪装问题。针对这些问题,提出一种利用两种背景差分法分别作用于颜色和深度数据的目标检测方法。在场景颜色图像中采用基于改进的局部二值相似性模式(LOBSTER)的背景差分法检测运动目标;在深度图中,对孔洞(无深度值)像素和有深度值的像素分别进行背景建模,得到一个混合的背景模型,然后用混合高斯模型背景差分法得到深度图的二值图像,再用差分后的Canny边缘检测图来补充目标的轮廓;将两种二值图像与Canny边缘检测的差分图用逻辑运算进行结合,形成运动目标。在后期处理中,加入了孔洞填充、形态学滤波及腐蚀膨胀操作,进一步提高目标的检测精度。实验结果表明,上述方法在测试数据集上得到的目标检测精度较高。

With the help of depth camera data,the color camouflage in object detection can be solved to some extent,but depth camouflage arises.To deal with these problems,we propose an object detection method based on two background subtraction methods which are respectively used for color data and depth data.The moving object is detected by the background subtraction method based on the improved local value of binary similarity model(LOBSTER)in color image of the scene.Background modeling is respectively applied to hole(no depth value)pixels and pixels with depth value to get a mixed background model.Then the binary images of the depth maps are obtained by the background subtraction method based on mixtures of Gaussians(MOG),and the contour of the object is supplemented by the difference maps obtained from Canny.Two binary images and Canny edge detection difference maps are combined using a logical operation to obtain the moving object.The holes filing,morphological filtering,corrosion and expansion operation are added in the post processing to further improve the object detection accuracy.The experiment shows that the proposed method can achieve higher object detection accuracy on the test data set.