首页|基于关键点匹配的泊车位检测算法

基于关键点匹配的泊车位检测算法

扫码查看
由于泊车位的类型多种多样,在复杂的环境下,智能车辆搭载的自动泊车系统可能因为无法准确识别泊车位而失效,造成驾驶员的操作体验感降低、事故风险率增加.针对车位线模糊、被遮挡和光照不足等情况下导致车位特征不明显的问题,提出了一种基于关键点匹配的泊车位检测算法,利用关键点热力图检测入口线的中心点和角点位置坐标;利用单位方向向量的亲和场匹配中心点到对应的角点;利用多区域的中心方向代表车位的边线方向;利用中线网格单元的占用情况来预测车位的占用状态;并设计了基于YOLOv8 的自校准卷积模型YSCCNet进行特征提取,可以实现包含位置、方向、占用情况等完整信息的泊车位检测.
Parking Slot Detection Algorithm Based on Keypoint Matching
Due to the diverse types of parking spaces,in more complex environments,the automatic parking system in-stalled on the smart vehicle may fail due to the inability to accurately detect the parking slot,resulting in a reduced driver's operating experience and an increased risk of accidents.This paper proposes a parking slot detection algorithm based on keypoint matching to address the problem of unclear parking space features caused by blurry parking lines,occlusion,and insufficient lighting.This paper uses keypoint heatmaps to detect the center and corner position coordinates of the entrance line;we utilize the affinity field of the unit direction vector to match the center point to the corresponding corner point.It represents the center direction of multiple regions as the edge direction of parking slots.This paper predicts the occupancy status of parking spaces by utilizing the occupancy of grid cells on the centerline.A self calibrated convolutional model YSCCNet based on YOLOv8 is designed for feature extraction,which can achieve parking slot detection with complete in-formation such as position,direction,and occupancy.

parking slot detectionkeypoint heatmapaffinity field matchingself-calibrated convolutional model

郑锐滔

展开 >

广东工业大学自动化学院,广东 广州 510006

泊车位检测 关键点热力图 亲和场匹配 自校准卷积模型

2024

工业控制计算机
中国计算机学会工业控制计算机专业委员会 江苏省计算技术研究所有限责任公司

工业控制计算机

影响因子:0.258
ISSN:1001-182X
年,卷(期):2024.37(12)