基于视觉的运动目标识别与跟踪系统研究

发布时间：2018-06-02 14:47

本文选题：循环移位 + 颜色特征　；参考：《哈尔滨工程大学》2016年硕士论文

【摘要】：运动目标跟踪是计算机视觉的重要研究领域,其在公共安全、武器制导、教育、医疗等方面具有广阔的应用前景。运动目标的识别跟踪是结合目标识别与跟踪的方法对图像序列中的运动目标进行检测,识别,跟踪的过程。近年来随着计算机视觉技术的发展,运动目标跟踪成为热点研究问题。目前虽已有较多出色的目标跟踪算法,但在实际应用中还面临着诸多影响跟踪的因素,如光照变化、运动目标形变、尺度变化、背景干扰、目标遮挡等,因此研究鲁棒的运动目标跟踪算法仍是一项具有重要意义的课题。目前学术领域已出现不少出色的跟踪算法,CSK跟踪算法便是其中之一,其优点在于利用循环移位的方法进行稠密采样并结合快速傅里叶变换进行分类器训练,跟踪速度出众。但其仅简单的使用目标的灰度特征,对目标外观描述能力不足,分类器的参数更新过于线性化无法适应目标发生较大变化时的跟踪。所以本文针对上述算法存在的不足,借鉴CN算法对CSK算法的改进思想,利用概率潜语意分析算法对运动目标颜色特征进行描述,同时采用PCA降维降低特征维度并去除特征冗余信息,对目标特征描述更为简练与鲁棒。在对分类器的训练中,在原有参数更新方式中引入权重参数β并对参数更新方案进行非线性化处理,使得分类器的训练与更新趋于稳定与准确。接着对benchmark测试平台的算法集与测试数据集进行扩充,利用扩充后的测试平台对改进后的算法与现有的效果出色的目标跟踪算法进行跟踪精度、成功率以及时间鲁棒性(TRE)、空间鲁棒性(SRE)的对比分析,同时对改进前后的算法也进行上述指标的对比分析。改进后的算法对很多具有复杂特征的序列都有较好的跟踪效果,如:光照变化、背景杂乱、遮挡、形变、运动模糊等。最后对改进后的算法进行实验验证,应用改进的目标跟踪算法,设计一个实验性的基于无人机的目标识别跟踪系统。以无人机作为控制目标,通过无线网络进行图像数据传输,数据传输模块进行控制指令的通信,跟踪地面运动小车,进行目标自动识别跟踪实验,并在实验过程中验证算法对跟踪过程中出现的干扰因素的处理,实验结果表明算法识别跟踪速度快,抗外界干扰能力强。
[Abstract]:Moving target tracking is an important research field in computer vision. It has a broad application prospect in public safety, weapon guidance, education, medical treatment and so on. The recognition and tracking of moving targets is the process of detecting, recognizing and tracking moving targets in the image sequence combined with the method of target recognition and tracking. In recent years, with the development of computer vision technology, moving target tracking has become a hot topic. Although there are many excellent target tracking algorithms, there are still many factors that affect tracking, such as illumination change, moving object deformation, scale change, background interference, target occlusion and so on. Therefore, the study of robust moving target tracking algorithm is still an important topic. At present, there are many excellent tracking algorithms in academic field, such as CSK tracking algorithm, which has the advantage of dense sampling with cyclic shift method and classifier training combined with fast Fourier transform (FFT), and the tracking speed is outstanding. However, it only uses the gray feature of the target simply, and its ability to describe the appearance of the target is insufficient, and the parameter update of the classifier is too linear to adapt to the tracking of the target when it changes greatly. Therefore, in view of the shortcomings of the above algorithms, this paper uses CN algorithm to improve the CSK algorithm and uses probabilistic latent semantic analysis algorithm to describe the color features of moving objects. At the same time, PCA is used to reduce the dimension of the feature and remove the redundant feature information, which is more concise and robust to the target feature description. In the training of classifier, the weight parameter 尾 is introduced into the original parameter updating mode and the parameter updating scheme is nonlinear, which makes the training and updating of classifier tend to be stable and accurate. Then, the algorithm set and test data set of benchmark test platform are extended, and the improved algorithm and the existing target tracking algorithm are used to track the precision of the improved algorithm. The success rate, time robustness and spatial robustness are compared and analyzed, and the improved algorithm is also compared with the above indexes. The improved algorithm has good tracking effect for many sequences with complex features, such as illumination variation, background clutter, occlusion, deformation, motion blur, etc. Finally, an experimental target recognition and tracking system based on UAV is designed by using the improved target tracking algorithm. The UAV is used as the control target, the image data is transmitted through wireless network, the data transmission module communicates the control instructions, tracks the ground motion vehicle, and carries on the target automatic recognition and tracking experiment. The experimental results show that the algorithm has high speed and strong ability to resist external interference.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】