| Multi-target tracking algorithms play a huge role in intelligent video surveillance,unmanned driving,and new retail user portraits.This research has broad market prospects and value.However,the current multi-target tracking field has problems such as target positioning,target occlusion and half occlusion,operating speed and lighting conditions.Therefore,this paper proposes to introduce a single target tracking algorithm based on a fully convolutional twin network into the field of multi-target tracking,and proposes an online intelligent monitoring scheme with the multi-target tracking algorithm as the core.First,this article introduces the SiamMask network and the target depth information feature.The single-target tracking network SiamMask network is applied to the field of multi-target tracking,and the target depth information is obtained through the structured light-based Realsense camera,and the input channel network structure of the SiamMask network is improved,so that the network can simultaneously obtain image RGB features and depth information features.This method effectively solves the problem of target positioning and occlusion,and avoids the impact of illumination to a certain extent.Then,this paper proposes a multi-target tracking algorithm fusing the full convolutional twin network and the ReID network.The ReID network is integrated with the fully convolutional tracking network,and the network is trained through multi-task learning.It has a good tracking effect in multi-target tracking,and the running speed is faster than most similar algorithms.It solves the problem of the multi-target tracking algorithm based on the SiamMask network,and the running speed drops sharply as the number of targets increases,and it meets the real-time tracking requirements.And applied to the design of monitoring scheme,to achieve online tracking,passenger flow statistics and online multi-target behavior identification and other functions. |
PDF Full Text Request