基于深度学习的视觉目标跟踪算法

A Visual Target Tracking Algorithm Based on Depth Learning

提出一种基于深度学习的视觉单目标跟踪方法.该方法在统一的网络框架下处理在线视觉单目标跟踪问题,在获取输入视频帧的基础上,通过结合贝叶斯损耗层的深度卷积神经网络来估计正负样本集的概率密度分布并计算出所有目标候选位置的得分,继而利用贝叶斯分类确认目标位置.针对正样本集个数有限的问题,该网络采用先对通用的目标特征进行预离线训练,然后通过多个步骤进行微调,在线微调主要是对每一帧中的图像外形特征进行学习.此外,该方法采用两级迭代算法自适应地更新网络参数并保持目标/非目标地区的概率密度.仿真实验表明了所提算法的有效性和可靠性.

A visual single target tracking method based on depth learning is proposed.The method under the unified network framework for processing the problem of online visual single target tracking,on the basis of taking the input video frames,by combining deep convolutional neural networks with Bayesian loss layer to estimate probability density distribution of positive and negative examples,and then calculate the score of all the candidate target position.In order to deal with the limited number of positive training examples,the network is pre-trained offline for a generic image feature representation and then is finetuned in multiple steps.An online fine-tuning step is carried out at every frame to learn the appearance of the target.In addition,the proposed method uses two stage iterative algorithm to adaptively update the network parameters and keep the probability density of the target/non target region.