基于Cauchy模型的行人轮廓提取及目标检测

Pedestrian profile extraction and target detection based on Cauchy model

摘要:

为了提高行人目标轮廓参量的提取精准度数值，实现对待监测目标的实时稳定跟踪，提出基于Cauchy模型的行人轮廓提取及目标检测算法。基于Cauchy分布原理，估计行人轮廓目标的最大似然值，再结合计算第二类统计量方法，完成基于Cauchy模型的行人目标统计建模。在此基础上，建立卷积神经网络，利用卷积化与反卷积参量，提取Gabor行人轮廓特征。在目标图像分割理论的作用下，识别既定区域内的所有行人目标，持续标记各类已存在的行人目标，实时检测行人轮廓目标，实现基于Cauchy模型行人轮廓提取及目标检测。实验结果表明，与Kinect型检测算法相比，应用Cauchy型算法后，行人目标轮廓的检测精度值提高至93%，而PTR实测指标降低至3.97，可有效实现待监测行人轮廓目标的实时稳定跟踪。

Abstract:

In order to improve the accuracy of pedestrian contour parameter extraction and realize the real-time and stable tracking of the target to be monitored, a pedestrian contour extraction and target detection algorithm based on Cauchy model is proposed. Based on the Cauchy distribution principle, the maximum likelihood value of pedestrian contour target is estimated, and then combined with the calculation of the second type of statistics method, the pedestrian target statistical modeling based on Cauchy model is completed. On this basis, a convolution neural network is established, and Gabor pedestrian contour features are extracted by convolution and deconvolution parameters. Under the function of target image segmentation theory, all pedestrian targets in the given area are identified, all kinds of existing pedestrian targets are continuously marked, and pedestrian contour targets are detected in real time, and pedestrian contour extraction and target detection based on Cauchy model are realized. The experimental results show that compared with Kinect detection algorithm, the detection accuracy of pedestrian contour is improved to 93% after using Cauchy algorithm, while the measured PTR index is reduced to 3.97, which can effectively realize the real-time and stable tracking of pedestrian contour target to be monitored.