基于子块运动补偿的运动目标检测

鱼眼相机成像视角大，获得信息丰富，在车载应用中具有广阔应用前景.本文提出了一种适用于移动单目鱼眼相机的运动目标检测方法.首先，提出一种子块运动补偿模型补偿图像背景运动，解决了现有运动补偿模型对强视差背景补偿效果不好的问题.其次，在子块运动补偿模型参数求解时，通过引入自车运动参数简化模型参数个数，并结合直接方法求解，避免了传统基于特征点匹配方法求解参数时易受误匹配特征点影响的问题.然后，针对鱼眼相机的成像形变问题，本文提出了一种三平面校正方法获取鱼眼图像的子块运动补偿图像.最后，利用鱼眼图像的子块运动补偿图像和真实拍摄图像的差异信息实现运动目标检测.多种测试场景下的实验结果表明了本文方法的有效性.

Moving Object Detection Based on Sub-Block Motion Compensation

With its wide-angle imaging and information richness,the fish-eye camera has a brilliant prospect in application.This paper presents a moving object detection method for on-board monocular fish-eye cameras.Firstly,a sub-block motion compensation model is proposed to compensate image background motion,which solves the ineffective strong parallax scene compensation problem.Secondly,when it comes to solve the parameters of sub-block motion compensation model,ego-vehicle motion parameters are introduced to simplify the number of model parameters,and the direct method is used to avoid the problem that the traditional feature-point-based matching method is susceptible to mismatching feature points.Then for the image distortion problem,this paper proposes a three-plane rectification method to obtain sub-block motion compensation images.Finally,the moving object detection is realized using the difference between sub-block motion compensation images and captured images.Experimental results show the effectiveness of the proposed method.