基于深度学习的多媒体移动物体检测技术研究

Research on Multimedia Mobile Object Detection Technology Based on Deep Learning

为了解决深度学习技术无法满足用户对多张照片移动目标物体识别的需求,提出基于深度学习的多媒体移动物体检测技术.提出基于前馈神经网络的目标定位技术,设计目标定位模型,计算目标边界框的大致位置以及概率;对视频中的目标行为进行预测;利用卷积神经网络融合特征和自然语言搜索特征完成自然目标搜索.

In order to solve the problem that deep learning technology can not meet the user's need for moving object recognition in multiple photos,a multimedia mobile object detection technology based on deep learning is proposed.Firstly,a target location technology based on feedforward neural network is proposed,and a target location model is designed to calculate the approximate position and probability of the target boundary box,as well as the complete algorithm flow design.Secondly,based on the method of time-space fusion,the feature extraction of time-space feature fusion and the extraction of time information from 3D convolution network are used to predict the target behavior in video.Finally,by using natural language and extracting local features,using convolutional neural network fusion features and natural language search features to complete natural target search.