一种融合检测、跟踪与分割的行人多任务算法

A PEDESTRIAN MULTI TASK ALGORITHM COMBINING DETECTION,TRACKING AND SEGMENTATION

随着图像处理算法需求的增长,单个视觉算法往往难以满足任务需求,针对该现象,提出一种同时具备检测、实例分割和多目标跟踪的多任务处理算法。算法采用Anchor Free的框架实现一阶段的检测、分割与跟踪,使用基于网格的预测策略降低多任务分支带来的计算量增加,降低了模型部署的算力需求。跟踪分支采用词嵌入的方式对跟踪对象进行编码,根据编码间的距离进行关联。分割分支采用Mask系数和原始掩膜组合的方式,平衡了算法运行的精度和速度。实验表明,该多任务算法在实时运行的基础上能够满足一定的精度需求。

With the increasing demand of image processing algorithm,a single vision algorithm is often difficult to meet the task requirements.To solve this problem,a multi task processing algorithm with detection,instance segmentation and multi-object tracking is proposed.The Anchor Free framework was used to realize one-stage detection,segmentation and tracking.The grid-based prediction strategy was used to reduce the computational load caused by multi task branching.The tracking branch encoded the tracking objects by word embedding,and associated them according to the distance between the word.The segmentation branch adopted the combination of mask coefficient and original mask to balance the accuracy and speed of the algorithm.The experimental results show that this algorithm can meet the accuracy requirements on the basis of real-time operation.