基于人工复眼的红外目标运动检测研究

Research on Infrared Target Motion Detection Based on Artificial Compound Eye

摘要: 红外光学成像应用于视觉目标运动检测具有可昼夜运用的特点。为了提升准确、可靠地检测物体(目标)运动方向的能力，提出一种红外目标运动方向检测的人工果蝇复眼模型(简称ACEM模型)。首先，对输入的红外视频进行帧差处理获取目标运动信息，模拟果蝇复眼视网膜层；其次，通过设计非线性自适应带通滤波器和仿生中心侧向抑制滤波器及算法，模拟果蝇复眼视叶结构的薄板神经节层初级视觉滤波功能；然后，模拟髓质神经节层跨层神经元T4和小叶神经节层跨层神经元T5对ON/OFF通道信号做出方向选择性响应；最后，模拟小叶板神经节层整合T4和T5对于方向的敏感响应，形成运动方向分量的检测输出。通过对杂乱视觉背景下拍摄的红外视频序列进行测试，验证了所提出的ACEM模型对于水平和垂直方向运动检测的有效性和鲁棒性。

Abstract: Infrared optical imaging can be used day and night in visual target motion detection. In order to improve the ability to accurately and reliably detect the movement direction of objects (targets), an artificial drosophila compound eye model (ACEM model for short) for infrared target movement direction detection was proposed. Firstly, the target movement information is obtained by frame difference processing of the infrared video input, and the retinula layer of drosophila compound eye is simulated. Secondly, In order to simulate the primary visual filtering function of the lamina ganglionic layer in drosophila compound eye, a nonlinear adaptive band pass filter and a bionic central lateral suppression filter and its algorithms were designed. Then, the trans lamellar neuronsT4 in medulla ganglion layer and the trans lamellar neurons T5 in lobula ganglion layer were simulated to respond directionally selectively to ON/OFF channel signals. Finally, by simulating the lobula plate ganglionic layer, to integrate the directional sensitive response of T4 and T5, form the detection output of motion direction component. By testing Infrared video sequences taken against a cluttered visual background, the effectiveness and robustness of the proposed ACEM model for horizontal and vertical directional motion detection are verified.