粗细粒度超像素行人目标分割算法

针对车载视觉行人目标分割由于复杂场景对行人目标的分割结果产生干扰而出现信息冗余以及错误分割的问题,提出一种粗细粒度超像素行人目标分割算法。该算法以Mask R-CNN作为粗粒度一次分割,将所得结果经Slic超像素细粒度二次分割,融合两次输出结果来提高现有图像目标的分割精度,为行人目标识别和跟踪提供有益先验感知信息。经仿真验证,该算法能够对复杂背景情况下的图像进行有效分割,MS COCO标准公开集测试结果与原有Mask R-CNN检测算法相比,mAP提高0.71%,为图像识别和计算机视觉系统完成精准的预处理,具有较强的工程应用价值。

Coarse-fine-grained Super-pixel Pedestrian Object Segmentation Algorithm

Aiming at the problem of information redundancy and error segmentation in the interference of pedestrian object segmentation in the process of vehicle vision pedestrian object segmentation, this paper proposes a coarse-fine-grained super-pixel pedestrian object segmentation algorithm. The algorithm uses Mask R-CNN as the coarse-grained one-time segmentation, and the obtained result is sub-divided by Slic super-pixel fine-grained method. The two results are combined to improve the segmentation accuracy of the existing image object, it is beneficial for pedestrian object recognition and tracking. The simulation results show that the coarse-fine-grained super-pixel pedestrian object segmentation algorithm can effectively segment the image in complex background conditions. The test result on the MS COCO standard open set is 0.71% higher than the original Mask R-CNN detection algorithm. The system provides accurate pre-processing parts and has strong engineering application value.