汽车紧固卡扣视觉多目标跟踪研究

Research on Visual Multiple-Object Tracking of Automobile Fastening Buckle

汽车配件生产需要工人一个个目测汽车紧固卡扣是否缺装,但这种人工目测识别效率非常低。针对这个问题,文章在YOLOv3-tiny网络中融合了空间和通道注意力机制,并用K-Means聚类找出最佳的Anchor Box。与YOLOv3相比,该算法检测准确率有显著提升,速度提高了3倍。最后用Lucas-Kanade光流法和延迟精准机制,实现了对汽车紧固卡扣的跟踪和识别。在几个视频序列实验中,文中算法的MOTA(Multiple-Object Tracking Accuracy)较其他算法提高了约(3.7~7.5)%。在200次实验中,汽车塑料组合件和紧固卡扣识别正确率达到了100%。

The production of auto parts requires workers to visually check whether the car fastening buckle is missing.This manual method is inefficient.To solve this problem,the space and channel attention mechanism is applied to YOLOv3-tiny net-work,which uses K-Means clustering to find the best Anchor Box.This algorithm has significantly improved detection accuracy in comparison with original YOLOv3 algorithm,and it’s speed has increased by 3 times.Finally,the Lucas-Kanade optical flow method and the Delay-Precision mechanism are used to achieve the tracking and recognition of the car fastening buckle.In several video sequence experiments,the MOTA(Multiple-Object Tracking Accuracy)of this algorithm is about(3.7~7.5)%higher than other algorithms.In 200 experiments,the recognition accuracy of automotive plastic components and fastening buck-les reaches 100%.