摘要
目标识别实验是数字图像处理实验的一个创新性实验项目,学生选课率较高。传统的目标识别算法不能识别目标的位置信息,而且识别率较低。在深度学习理论构架下,设计了基于NMS及帧间差分的目标识别实验仿真算法,将帧间差分法融入识别过程,采用帧间差分法提取待识别视频的动态信息作为补充,增强候选框区域分割图像,并通过NMS算法对候选框进行筛选,提高识别率。仿真结果表明,算法识别出目标种类的同时能对目标在图像中的位置进行精确标定,并可以判断目标是否处于运动中,具有较高的识别率。
Target recognition experiment is an innovative experimental project of digital image processing experiment,and the rate of students choosing courses is high.The traditional target recognition algorithm can not recognize the target position information,and the recognition rate is low.Under the deep learning theory framework,design the target recognition based on the NMS and inter-frame difference experiment simulation algorithm,the inter-frame difference method into the identification process,using inter-frame difference method to extract the dynamic information to identify the video as a supplement,enhance the region proposal segmentation image,and through the NMS algorithm to filter region proposal,improve the recognition rate.Simulation results show that the algorithm can accurately calibrate the position of the target in the image and judge whether the target is in motion while recognizing the type of target,and has a high recognition rate.
作者
王辉
于立君
刘朝达
高天禹
WANG Hui;YU Lijun;LIU Chaoda;GAO Tianyu(College of Automation,Harbin Engineering University,Harbin 150001,China)
出处
《实验室科学》
2021年第1期52-56,共5页
Laboratory Science
基金
黑龙江省教改项目(项目编号:SJGY20180090
SJGY20180089)
哈尔滨工程大学教改项目(项目编号:JG2018Y06)
哈尔滨工程大学研究生教改项目(项目编号:JG2019Y09)。
