Description: 背景建模是实现运动目标检测与跟踪的关键技术之一。在实时视频监控系统中，对背景建模算法的运行时间及所提取出的背景图像的实时性有很高的要求，针对这一问题，提出了一种基于切比雪夫不等式的自适应阈值背景建模算法。算法利用切比雪夫不等式计算像素点色度变化的概率估计值，提出了一种自适应阈值分类方法，它将像素点快速分类为前景点、背景点及可疑点，再利用核密度估计方法对可疑点进行进一步分类，最后利用背景更新算法提取实时背景图像。实验结果证明，该算法能快速有效地区分特征明显的背景点与前景点，提高了背景图像提取的速度，对可疑点利用核密度估计方法降低了背景分割的误差，背景建模效果理想，运算速度快，适用于实时视频监控系统。-Background modeling is a key technology to realize the moving target detection and tracking. In real-time video surveillance system, there are high demands on uptime and background modeling algorithm is proposed to remove the background image in real time, for this problem, a Chebyshev inequality based on adaptive threshold background modeling algorithm. Cut algorithm uses to calculate the probability of Chebyshev inequality pixel color change estimates, an adaptive threshold classification method, it will be classified as pre-fast pixel of interest, background points and suspicious points, re-use kernel density estimation method suspicious point for further classification. Finally, background updating algorithm to extract real-time background image. Experimental results show that the algorithm can quickly and efficiently in the background of significant features of the region of interest with the previous point, improving the speed of extraction of the background image, the point of s Platform: |
Size: 1972224 |
Author: |
Hits:
Description: 时空上下文视觉跟踪（STC）算法的解读与代码复现
该论文提出一种简单却非常有效的视觉跟踪方法。更迷人的一点是，它速度很快，原作者实现的Matlab代码在i7的电脑上达到350fps。
该论文的关键点是对时空上下文（Spatio-Temporal Context）信息的利用。主要思想是通过贝叶斯框架对要跟踪的目标和它的局部上下文区域的时空关系进行建模，得到目标和其周围区域低级特征的统计相关性。然后综合这一时空关系和生物视觉系统上的focus of attention特性来评估新的一帧中目标出现位置的置信图，置信最大的位置就是我们得到的新的一帧的目标位置。另外，时空模型的学习和目标的检测都是通过FFT（傅里叶变换）来实现，所以学习和检测的速度都比较快。-Space-time visual tracking context (STC) algorithm for interpretation and code reuse the existing paper proposes a simple but very effective visual tracking method. More attractive is that it is fast, Matlab codes to achieve the original author reaches 350fps on i7 computer. The key point of the paper is a space-time context (Spatio-Temporal Context) access to information. The main idea is to be tracked through a Bayesian framework of goals and temporal relationship between its local context area is modeled to obtain objective and its surrounding area statistical correlation between low-level features. Then focus on the relationship between the biological and the integrated vision system to uate the spatial and temporal characteristics of attention of a new target position occurs confidence map, is the new position of maximum confidence of an objective position we get. In addition, the study and detection of target space-time model through FFT (Fourier transform) to achieve, so learni Platform: |
Size: 7207936 |
Author:老王 |
Hits:
Search - Detection and Tracking of Point Features