基于视觉压缩感知的传感网络行人目标辨识方法

Visual compressive sensing-based pedestrian identification in sensor networks

行人目标辨识是指在视觉传感网络中识别检测到的目标,对智能安防具有重要意义。对行人目标辨识所需数据进行压缩可提高视觉传感网络行人目标辨识的实时性。提出了一种基于视觉压缩感知的传感网络行人目标辨识方法。无线视觉节点获取行人目标图像后,首先提取图像中行人脸部的尺度不变特征,并采用特征字典对目标进行稀疏表示,得到目标特征直方图。然后视觉节点应用压缩感知方法对特征直方图进行数据压缩,并传输至中心节点。最后,中心节点应用非负正交匹配追踪算法重构特征直方图,并采用支持向量机对特征直方图进行分类辨识。实验表明,该方法能够在不影响行人目标辨识准确率的前提下,有效减少在视觉传感网络中进行行人目标辨识时所需传输的数据量。

Pedestrian identification is to identify a detected target emerged in visual sensor networks, which is important for the applications of intelligence security. Data compression for the pedestrian identification can enhance the instantaneity. This paper proposes a visual compressive sensing based pedestrian identification scheme in sensor networks. In this scheme, scale-invariant features are extracted and represented by a sparse feature histogram via feature dictionary. After that, the histogram is compressed according to compressive sensing theory and sent by the node to the server. At the server, a nonnegative compressive sampling matching pursuit algorithm is developed to reconstruct the compressed histogram. And then support vector machine is applied to classify the histogram to identify the person. The proposed scheme is verified with built visual sensor networks. The results show this scheme can greatly reduce the data amount needed to transmit without reducing the pedestrian identification accuracy.