改进Mask RCNN算法及其在行人实例分割中的应用

Mask RCNN算法在特征提取过程中存在语义信息丢失的问题，而自然场景中的行人具有姿态不同、遮挡和背景复杂等特点，导致算法应用于行人实例分割时检测准确性较差。对此，提出一种改进的Mask RCNN算法。在Mask RCNN网络的Mask分支中增加串联特征金字塔网络（CFPN）模块，对网络生成的多层特征进行融合，充分利用不同特征层的语义信息。在此基础上，执行RoI Align操作生成行人掩膜。仿照COCO数据集，从生活场景中拍摄1 000张图片，自建一个新的行人数据集。基于该数据集的实验结果表明，改进算法较原算法具有更高的检测精确率。

Improved Mask RCNN Algorithm and Its Application in Pedestrian Instance Segmentation

In the process of feature extraction, Mask RCNN algorithm will lose semantic information.However, pedestrians in natural scenes have different posture, occlusion and complex background, which lead to poor detection accuracy in pedestrian segmentation.To solve this problem, an improved Mask RCNN algorithm is proposed.In the Mask branch of the Mask RCNN network, the Concatenated Feature Pyramid Network(CFPN) is added to fuse the multi-layer features generated by the network, so as to make full use of the semantic information of different feature layers.On this basis, the RoI Align operation is performed to generate a pedestrian mask.Following the COCO data set, 1 000 pictures are taken from life scenes, and a new pedestrian data set is built.Experimental results based on the data set show that the improved algorithm has higher detection accuracy than the original algorithm.