基于卷积神经网络检测的单镜头多目标跟踪算法

Single-lens Multi-target Tracking Algorithm Based on Convolution Neural Network

目前卷积神经网络在图像识别分类领域已经取得了长足的进步,随着网络模型的优化改进,目标识别的准确度和帧速率都在提升,因此利用相对成熟的卷积神经网络模型进行多目标跟踪任务变得可行。论文即是利用卷积神经网络模型Fast R-CNN设计出多目标跟踪的算法,以Fast R-CNN作为模型检测的主框架,将训练样本分为目标和背景两个类并以此完成检测模型的线下训练,以对每个视频帧检测的方式完成整个视频的跟踪任务。实验结果表明,这种算法在实时性、准确性和鲁棒性上均有不错的表现。

At present,convolution neural network has made great progress in the field of image recognition and classification.With the optimization of the network model,the accuracy and frame rate of target recognition are improved. So the use of relativelymature convolution neural network model for multi-target tracking task becomes feasible. In this paper,the convolution neural net-work model Fast R-CNN is used to design a multi-target tracking algorithm. Taking the Fast R-CNN as the main frame of the modeldetection,the training samples are divided into two categories:target and background,therefore complete the offline training of thetest model,and complete the tracking task for the entire video in a manner that detects each video frame. The experimental resultsshow that this algorithm has good performance in terms of real-time,accuracy and robustness.