摘要： 视频监控的广泛应用使运动对象检测成为研究热点，但运动的不确定性增加了检测难度。鉴于人类视觉系统能高效地感知运动对象，研究者从神经生理学和心理学的角度提出了运动检测的生物学模型。根据上述研究成果，提出模拟初级视皮层的运动对象检测模型。使用三维Gabor时空滤波器模拟人类初级视皮层中简单细胞的经典感受野，通过非线性组合获取复杂细胞对运动对象刺激响应的运动能量，应用细胞的中心环绕作用及相关性运动检测增强运动信息并抑制环境干扰，采用信息融合获取运动对象的显著性图，并利用WTA神经网络模型实现对运动目标的感知。实验结果表明，该模型能有效检测到视频中的运动目标，运算速度较其他仿视神经加工的视觉注意模型更快。

关键词: 运动对象检测, 时空显著性, Gabor滤波器, 神经网络, 初级视皮层

Abstract: With the wide application of video surveillance, motion detection becomes a research focus. However, it becomes more difficult due to a lot of uncertainty in motion. Based on the human visual system more efficiently to perceive moving objects, researchers construct the model of motion perception based on the view of physiology and psychology. According to these studies, this paper proposes a moving object detection model simulating primary visual cortex. In this model, Classical Receptive Field(CRF) of simple cells in primary visual cortex is simulated by the three-dimensional Gabor spatial-temporal filter, and the motion energy as complex cells response is obtained by a nonlinear combination. Two characteristics: center-surround of cells and correlation motion detection is used to enhance motion information and suppress environment interference, and more information is fused to obtain a saliency map of moving object. WTA neural network model is used to achieve the perception of motion object. Experimental results show that the model can effectively detect the moving object in the video, and compute faster than other visual attention models based on visual neurons processing.

Key words: motion object detection, spatial-temporal saliency, Gabor filter, neural network, primary visual cortex

中图分类号: 

• TP18