摘要
车位线识别是自动泊车系统中感知环节的关键,本文基于轻量化的目标识别网络YOLOv2-Tiny实现车位角点检测,并通过对网络识别出的角点块区域进行灰度化、自适应二值化、开运算等预处理,后续进行骨架提取,利用概率霍夫变换检测角点骨架直线从而计算校正后的角点中心,并确定车位摆向。所提出的方法可用于检测不固定角度的平行、垂直与斜列车位,识别效果优异,检测效率高,可满足实时性要求。
Parking slot detection is vital for the perception part in Automatic Parking Assist(APA)system.The proposed algorithm achieved parking slot corner detection based on the light-weight target recognition network YOLOv2-Tiny,and the corner block identified by the network was pretreated by grayscale,adaptive binarization and opening operation.Then skeleton extraction and Progressive Probabilistic Hough Transformation(PPHT)were adopted to detect the slot line of the corner point to calculate the corrected corner point center,and further determine the position of the parking slot.The proposed method can be used to detect parallel,vertical and diagonal slots with unfixed angles,it performs satisfactory effect and high detection efficiency,which can meet real-time requirements.
作者
何俏君
郭继舜
关倩仪
钟斌
付颖
谷俊
HE Qiao-jun;GUO Ji-shun;GUAN Qian-yi;ZHONG Bin;FU Ying;GU Jun(GAC Automotive Research&Development Center,Guangzhou 511434,China)
出处
《汽车电器》
2020年第9期1-5,共5页
Auto Electric Parts
关键词
轻量化网络
角点检测
车位线识别
骨架提取
自动泊车
light-weight network
corner detection
parking slot recognition
skeleton extraction
Automatic Parking Assist