面向行人检测的异构视觉传感网络自适应标定

Self-adaptive calibration of hybrid visual sensor networks for pedestrian detection

异构视觉传感网络的标定是实现准确行人检测、跟踪、识别以及行为分析的关键和前提,对公共安全和智能安防具有重要意义。本文提出了一种面向行人检测的异构视觉传感网络自适应标定方法。首先,构建主动旋转变焦(PTZ)传感节点变焦成像模型,采用基于SURF特征点匹配的方法实现内参数自标定。其次,构建外参数分布式标定模型,将异构视觉传感网络外参数标定分解为节点局部标定和网络全局标定,从而避免采用单一中心节点进行集中式处理,提高方法的可扩展性;最后,为了提高标定精度,采用自适应混沌微粒群优化算法最小化内、外参数估计的再投影误差和轨迹匹配误差。实验结果表明,本方法有效降低了网络通讯量和标定误差,提高了异构视觉传感网络行人检测精度,具有重要的理论和实际应用价值。

The calibration of hybrid visual sensor networks（ HVSNs） is the key and premise to realize precise pedestrian detection,tracking,recognition and activity analysis,which has great significance to public security and intelligent security. This paper proposes a self-adaptive calibration method of hybrid visual sensor networks for pedestrian detection. First,a zoom imaging model for a PTZ camera sensor node is established. The intrinsic parameter self-calibration of the PTZ camera is realized adopting SURF based feature matching.Second,aiming at improving the fault tolerance and scalability,the extrinsic parameter distributed calibration model of the HVSNs is built,which decomposes the extrinsic parameter calibration of the HVSNs into the local calibration of visual sensor nodes and global calibration of the HVSNs. Thus,the centralized processing with single center node is avoided,and the scalability of the method is enhanced. Finally,in order to improve the calibration accuracy,an adaptive chaotic particle swarm optimization algorithm is applied to minimize the re-projection error and trajectory matching error of the intrinsic and extrinsic parameter estimation. The experiment results show that this method could effectively reduce the network traffic and calibration error,improve the pedestrian detection accuracy in HVSNs and has great theoretical and application value.