摘要
为解决遮挡、姿态变化等局部变化引起的行人检测性能下降问题,提出一种融合全局和局部特征的行人检测方法。首先,将人体分为全局和局部6个部件;然后,改进Haar-like特征描述子,用于快速提取人体局部部件特征,再融合全局部件的方向梯度直方图特征,构建人体的联合部件特征。最后,结合增强学习思路改进支持向量机学习方法,对联合部件特征进行训练和分类。实验结果表明,该方法正确率高,虚警率低,受遮挡、姿态变化影响小。
For solving the problem of performance degradation of pedestrian detection caused by local changes such as occlusion and pose variation, a pedestrian detection method fusing global and local features is proposed. First,it divides human body into six components including global and local parts.Then,it improves the Haar-like descriptor for fast extracting features of local components of human body, and it fuses the histogram of oriented gradients features of global component to build the joint components features of the human body. Finally, it improves the learning method of support vector machines by combined with boost learning idea,to train and classify the joint components features.Experimental results shows that, this method has high accuracy rate and low false alarm rate, and less influenced by occlusion and pose variation.
出处
《电子技术应用》
北大核心
2017年第4期133-137,共5页
Application of Electronic Technique