基于生物视觉的碰撞预警传感器

Collision warning sensor based on biological vision

对拥有碰撞预警能力的小叶大运动侦察器(LGMD)神经网络进行化简,以参数化方式设计电路,网络参数可以通过串口在线配置。以此为核心设计了碰撞预警传感器,FPGA综合结果表明像素处理时钟最高可达178MHz。设计了专用测试系统进行验证,在实时模式下碰撞预警传感器根据摄像头输入正确进行碰撞预警;测试模式下用视频序列测试碰撞预警传感器,其输出膜电势与软件仿真一致,表明碰撞预警传感器继承相关软件工作的碰撞预警特性。测试系统也作为硬件加速器,加快LGMD算法运行。

With collision warning capability, lobula giant movement detector （LGMD） is simplified to design circuit in parametric method, and network parameters can be configured online by RS-232. A collision warning sensor is designed based on LGMD module and from FPGA synthesis report, the pixel frequency is as much as 178 MHz. A test system are also designed for verification. Under real-time mode, the sensor can generate warning alarm correctly according to inputs from camera; while under test mode, use video sequence to test sensor, output membrane potential consists with software simulation, which demonstrates the designed sensor earries on collision warning characteristics of the previous related work in software. The test system ean also be used as hardware aceelerator to boost LGMD algorithm.