基于群智感知的街景变化检测方法

移动群智感知数据包含的图像和时空情境信息可用于检测街景图像变化，但是群智感知数据通常是低质和不规范的。为了准确检测街景发生的变化，主要解决由拍摄视角差异引起的数据低质问题。首先，针对大视差问题采用图像配准方法初步对齐图像并提取出配准特征点；然后，基于配准特征点分布从图像中提取感兴趣区域；随后，针对差值图像的误检内容，提出基于面积和多特征点的筛选法去除误检区域；最后，结合边缘检测和超像素分割算法提取完整的变化对象。与MDFNet方法进行比较，实验结果显示：当街景发生变化时，该方法的F1-measure值为55.8%，增长6%，错误率为10.8%，降低24%；当街景无变化时，该方法的错误率为2.8%，下降28%。

Change detection method of street view image based on crowdsensing

The images and spatiotemporal contextual information contained in mobile crowdsensing data can be used to detect changes in street view images, but crowdsensing data is usually low-quality and non-standard. In order to accurately detect the changes in the street view, this paper mainly solved the problem caused by low-quality datum which had been collected from different shooting perspectives. Firstly, for the problem of large parallax, it used the image registration method to initially align the images and extract the registration feature points. Then, it extracted the region of interest from the image based on the distribution of the registration feature points. Thirdly, aiming at the error detection content of the difference image, it proposed a screening method based on area and multi-feature points to remove the error detection region. Finally, it extracted the complete changing object by combining edge detection and superpixel segmentation algorithm. Compared with MDFNet method, when the image pairs change, the F1-measure value and error rate of this method are 55.8% and 10.8%, and they have increased by 6% and decreased by 24% respectively. When there are no changes, the error rate of this method is only 2.8% and it is reduced by about 28%.