摘要
运动目标检测是物体跟踪和行为分析等高级任务的重要基础。目前,传统的背景减除算法和基于深度学习的目标检测算法都不能同时保证高灵活性和高准确性。提出了一种基于深度学习的运动目标检测后处理方式,通过融合当前视频帧图片和传统背景减除算法的检测结果图,重新对卷积神经网络进行训练,以提高现有的运动目标检测算法对复杂场景的处理能力。通过在CDNet2014数据集的实验验证了该后处理方式可以提升大多数的传统的背景减除算法的检测性能。
Detection of moving objects is an essential foundation for high-level tasks such as object tracking and behaviour analysis. Currently, both traditional background subtraction algorithms and deep learning-based moving detection algorithms cannot guarantee high flexibility and high accuracy at the same time.In this paper, a post-processing approach based on deep learning is proposed to improve the existing moving objects detection algorithms for complex scenes by fusing current video frames and detection results of traditional background subtraction algorithms and retraining the convolutional neural network. The experiments on CDNet2014 dataset can verify that the post-processing approach proposed in this paper can improve the detection performance of almost all traditional background subtraction algorithms.
出处
《工业控制计算机》
2022年第10期106-108,共3页
Industrial Control Computer
关键词
运动目标检测
深度学习
背景减除算法
卷积神经网络
后处理
moving objects detection
deep learning
background subtraction algorithm
convolutional neural network
post-process