Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (15): 169-183.DOI: 10.3778/j.issn.1002-8331.2203-0032

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Neural Network for Moving Small Target Pedestrian Detection Based on Episodic Memory

ZHANG Benkang, HU Bin   

  1. School of Computer Science and Technology, Guizhou University, Guiyang 550025, China
  • Online:2022-08-01 Published:2022-08-01

基于情景记忆的运动小目标行人检测神经网络

张本康,胡滨   

  1. 贵州大学 计算机科学与技术学院,贵阳 550025

Abstract: Reliable detection of small target pedestrians from visual scenes is a critical basis for building future intelligent video surveillance systems. However, the tiny sizes and blurred textures of moving small targets make it difficult for these existing traditional methods of pedestrian target detection to cope. To address this problem, the paper proposes an artificial visual neural network(STPDNN) model for moving small target pedestrian detection, by means of the neural structure properties of locust vision systems and the episodic memory cognition mechanism of the medial temporal lobe(MTL) inhuman brains. The proposed neural network consists of two parts: presynaptic and postsynaptic sub-networks. Therein, the presynaptic network obtains visual motion cues characterizing the low-order features of targets by simulating the neural processing mechanisms of locust visual systems; the postsynaptic network extracts the episodic memory higher-order information of pedestrian targets from low-order visual signals to response selectively motion targets. The results of systematic experiments show that the proposed STPDNN can effectively detect moving small target pedestrians in different visual scenes. This research work involves dynamic visual information processing of pedestrian targets inspired by biological optic nerve mechanisms, which can provide novel ideas andcreative methods for pedestrian detection and moving behavior analysis in future intelligent video surveillance.

Key words: moving target detection, small target-pedestrian, episodic memory, locust vision nerve, intelligent video surveillance

摘要: 从视觉场景中可靠地检测小目标行人对象是构建未来人工智能视觉系统的重要基础。由于运动小目标的视感尺寸小且纹理特征模糊,导致现有的传统行人目标检测方法难以应对。针对该问题,基于蝗虫视觉系统的神经结构特性,借助人类大脑内侧颞叶(MTL)情景记忆认知机理,提出一种适用于运动小目标行人检测的人工视觉神经网络(STPDNN)模型。所提出的神经网络包括两部分:突触前和突触后子网络。其中,突触前网络模拟蝗虫视觉系统加工处理视觉信号的神经机理,获得表征目标对象低阶特征的视觉运动线索;突触后网络从低阶视觉信号中提取出行人目标的情景记忆高阶信息,以实现对运动目标的偏好性响应。系统性的实验结果表明,提出的STPDNN可有效检测视觉场景中的运动小目标行人对象。该研究工作涉及生物视神经机理启发的行人目标动态视觉信息加工处理,可为智能视频监控中的行人检测识别与运动行为分析提供新思想、新方法。

关键词: 运动目标检测, 小目标行人, 情景记忆, 蝗虫视觉神经, 智能视频监控