运动目标检测和跟踪是智能视频监控的基础和前提，其在交通、军事、工业以及医学等各个领域具有广泛的应用前景。运动目标检测是指在视频图像中判断是否有前景目标的运动，如果有运动目标的存在，则对其进行初始定位。运动目标检测性能的好坏直接影响到后续跟踪的准确性和有效性。因此，本文在背景建模的基础上对运动目标检测与跟踪算法进行了深入研究。 本文首先研究了处于变速和光照突变情况下的目标检测算法，分析了处于变速运动情况下的目标尺寸、速度和停留时间与背景学习率之间的关系，提出了分区域采用自适应学习率更新背景模型的改进方法。针对监控场景中存在光照突变和阴影区域的情况，本文在结合双阈值Ostu分割方法和阴影抑制模型的基础上，利用三帧差分法对光照不敏感的特性，提出改进的三帧差分算法。经过实验，结果表明改进算法能够完整、准确地检测出运动目标的轮廓，具有较好的鲁棒性。 其次研究了传统的基于Mean Shift算法的目标跟踪问题。针对传统Mean Shift算法在复杂背景环境中容易造成跟踪失败的情况，本文提出了改进的Mean Shift跟踪算法。该算法以目标的质心和LBP纹理作为特征值。经过实验，结果表明改进的跟踪算法有效地改善了传统Mean Shift跟踪算法中丢失像素点所在空间位置信息的不足，提高了算法的稳定性和鲁棒性。

Moving object detection and tracking are the basis and prerequisite for intelligent video surveillance, and it can be applied widely in fields of transportation, military, industry and medicine. Moving object detection is a detecting process which refers to judge whether there is a foreground object in video image movement, and then locate its initial position if foreground object exists. The performance of moving object detection has a direct impact on the tracking accuracy and effectiveness of the follow-up process. Therefore, we have made a in-depth research on the moving object detection and tracking algorithm on based of background modeling. Firstly we made a study of the detection methods which object is under the situation of variable speed and light mutation. In the case of moving object is in a variable motion, we will analyze the relationship between the moving object size, speed, dwell time and the background learning rate and propose an improving measure which adopt different learning rate to update background model in different areas. When there have the light mutation and shadow areas exist in the video scene, then we use the characteristics of three-image difference which is not sensitive to the the light mutation, combining with dual-threshold segmentation method and shadow suppression model, and put forward a modified three-image difference algorithm. By the experiment, the results have showed that the improved algorithm can completely and accurately detect the contour of the moving object with the robustness. Secondly, we also made a study the traditional object tracking algorithm based on the Mean Shift algorithm. For the issue that the traditional Mean Shift algorithm is likely to cause the tracking failure in complex background environment, this paper proposes an improved tracking algorithm. The algorithm takes the centroid of the object and the LBP texture as the characteristic value. By the experiments, the results have showed that the improved tracking algorithm can effectively improve the shortcoming of missing pixel information in spatial position in traditional Mean Shift tracking algorithm, which can improve the stability and robustness of the algorithm.