基于视频序列的运动目标检测与跟踪算法研究

发布时间：2018-12-18 20:07

【摘要】：视频目标跟踪在计算机视觉领域中有着十分重要的地位,在智能交通、公共安全、人工智能等多方面拥有广阔的应用前景。但是传统的目标跟踪算法存在许多问题,例如受环境的影响比较大,当运动目标出现遮挡时容易出现跟踪丢失无法重新捕获目标。如何对运动目标进行有效、准确的跟踪是计算机视觉领域中一直关注的问题。本文首先研究了基于Codebook算法的运动目标检测算法,给出了算法的基本原理和性能特点。针对原始Codebook算法在运动目标检测中计算速度比较慢的问题,本文提出了一种基于颜色空间改进和参数优化的Codebook运动目标检测算法。将原Codebook算法从RGB空间转换到YUV空间,并且通过分析将3个颜色空间通道减少到1个,同时用亮度差值代替原始最大、最小亮度参数,通过引入码字权重系数来删减和优化其他参数,从而使得优化后的Codebook算法在运动目标检测中能够在具有较高准确性的情况下也较大程度提高算法的计算速度。针对原始TLD算法在跟踪过程中容易出现跟踪漂移的问题,本文提出了一种基于关键特征点的TLD算法简称为STLD,采用包含丰富信息量的特征点来代替原TLD算法中的Grid均匀采样,提高了运动目标的特征采样点的跟踪准确度,抑制了原TLD算法跟踪漂移,同时也减少了采样点的跟踪丢失率,因此具有更好的漂移抑制效果和更快的运算速度,提高了算法的鲁棒性。针对原始TLD算法在运动目标出现遮挡或者发生形变时会导致跟踪失败的问题,本文提出一种基于Kalman滤波的特征点TLD算法简称为KSTLD,在STLD算法检测器前段引入预测器,通过Kalman预测器对目标位置进行预测,加强视频序列前后帧中运动目标位置的相关性。从而得到当前视频图像中运动目标所在位置的大致区域,该预测器结果和STLD算法检测器中三级分类器进行结合,改善了在遮挡环境下,对运动目标的检测效果,提高了 STLD算法检测器的准确性和运算速度。
[Abstract]:Video target tracking plays an important role in the field of computer vision, and has a broad application prospect in intelligent transportation, public safety, artificial intelligence and so on. However, there are many problems in the traditional target tracking algorithm, such as being influenced by the environment. When the moving target is blocked, the tracking loss is easy to occur and the target can not be captured again. How to track moving targets effectively and accurately is always concerned in the field of computer vision. In this paper, the moving target detection algorithm based on Codebook algorithm is studied, and the basic principle and performance characteristics of the algorithm are given. In order to solve the problem of slow computing speed of the original Codebook algorithm in moving target detection, this paper proposes a new Codebook moving target detection algorithm based on color space improvement and parameter optimization. The original Codebook algorithm is transformed from RGB space to YUV space, and three color space channels are reduced to one by analyzing, and the original maximum and minimum luminance parameters are replaced by luminance difference. By introducing the codeword weight coefficient to delete and optimize other parameters, the optimized Codebook algorithm can improve the computational speed of the algorithm in the case of high accuracy. In order to solve the problem that the original TLD algorithm is prone to trace drift in the tracking process, a TLD algorithm based on the key feature points is proposed in this paper. In short, STLD, uses feature points with abundant information to replace the Grid uniform sampling in the original TLD algorithm. The tracking accuracy of the feature sampling points of moving targets is improved, and the original TLD algorithm is restrained, and the tracking loss rate of the sample points is also reduced, so it has better drift suppression effect and faster operation speed. The robustness of the algorithm is improved. In order to solve the problem that the original TLD algorithm can cause tracking failure when the moving object is occluded or deformed, a feature point TLD algorithm based on Kalman filter is proposed in this paper, which is referred to as KSTLD, introducing predictor in front of STLD algorithm detector. The target position is predicted by Kalman predictor to enhance the correlation between the moving target position in the frame before and after the video sequence. The result of the predictor is combined with the three-level classifier in the STLD algorithm detector to improve the detection effect of moving target in the shaded environment. The accuracy and speed of STLD detector are improved.
【学位授予单位】：扬州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】