联合CRF和可变部位模型的行人检测方法

Combining CRF and Deformable Part Model for Pedestrian Detection

行人目标检测在许多领域有着广泛的应用,它是计算机视觉研究的焦点之一。基于部位的检测方法在行人检测方面表现出非常出色的性能,在人体姿态变化方面具有很强的适应性,但是对于部位遮挡问题效果不佳。当判别阈值较高的时候,漏检率很高。考虑LSVM方法对遮挡信息挖掘不足,在可变部位模型的基础上,针对部位遮挡问题,建立了条件随机场模型,采用两层分类器。在参数学习中,采用随机梯度下降和置信传播算法优化条件随机场的目标函数。实验结果表明,该文提出的方法在遮挡问题方面表现出较好的效果。

Pedestrian detection has been widely used in many fields. It is one of the focus in computer vision. The part-based detection method in the pedestrian detection shows excellent performance and has a strong adaptability in posture change of human body. But it is not good for Occlusion problem. When the Discriminative threshold is higher, miss rate is very high. Considering the disadvantage of LSVM method for mining hidden information, a two layers classifier was proposed based on the deformable parts model establishing conditional random field model for Occlusion problem. For learning model parameters, the stochastic gradient descent and belief propagation algorithm optimization objective function of the random field conditions were used. The experimental results show that the proposed approach achieves good effect for Occlusion problem.