基于多任务目标检测的条形码倾斜矫正算法研究

条形码识别技术在日常工作中发挥着巨大作用,尤其是在智能物流包裹分拣领域。该技术主要分为三个部分:条形码检测、矫正和译码。目前条形码检测和译码技术较为成熟,而在条形码倾斜矫正技术上研究效果一般。为提升条形码矫正效果,设计一种矫正算法。先对条形码倾斜程度进行分类,再进行角度回归,有效降低条形码矫正任务难度;并将该算法与单阶段检测器融合构成多任务目标检测算法,协同促进发挥检测和矫正的作用。实验表明:余弦距离角度损失函数更加适合角度回归任务,针对条形码倾斜程度分类有助于提升条形码矫正效果。与其他相关算法对比,该算法在矫正准确率、实际译码率和速度上均取得最优的效果。

BARCODE TILT CORRECTION ALGORITHM BASED ON MULTI-TASK OBJECT DETECTION

Barcode recognition plays a huge role in daily work,especially in the field of intelligent logistics parcel sorting.It is mainly divided into three parts: barcode detection,correction and decoding.At present,barcode detection and decoding technology is relatively mature,however,barcode tilt correction research indicates a flat effect.In order to improve the barcode correction effect,a correction algorithm was designed.By classifying the tilt degree of the barcode and then performing angle regression,the difficulty of the barcode correction task was effectively reduced,and the algorithm was combined with the one-stage detector to form a multi-task object detection.The algorithm synergistically promoted the role of detection and correction.The experiments show that the cosine distance angle loss function is more suitable for the angle regression task.The classification of the barcode tilt degree can improve the barcode correction effect.Compared with other related algorithms,the proposed algorithm achieves the best results in correcting accuracy,decoding rate and speed.