Research On Person Re-identification Based On Combination Of Spatial-temporal Features And Appearance Features

In recent years,because people pay more attention to security,and the continuous upgrading of monitoring technology and storage equipment,public areas are covered by large monitoring equipment.However,it is impossible to deploy a camera network covering the entire public scene.Therefore,there are a large number of invisible areas.This requires a method that can connect the information obtained from the covered area.Due to the objective reasons such as different camera angle and light intensity,the storage device obtains a large number of low-quality pictures.Based on the above analysis,this paper use spatial-temporal information to assist pedestrian recognition.The main research work of this paper is as follows:1.To solve the problem of pedestrian appearance similarity,this paper proposes a hybrid Gaussian spatial-temporal model constraint method,which counts the walking time difference of pedestrians passing through different cameras,excavates potential pedestrian label information,and fits the real walking time distribution of pedestrians.Through the weight fusion of appearance measure matrix and space-time distance matrix,we can filter the impossible pedestrians and constrain the association between groups of pedestrians.2.Aiming at the low quality problem of pedestrian image in the actual scene,this paper uses the image resolution to enhance the color information of the image itself and improve the signal-to-noise ratio of the image.For the interference of different pedestrians with similar walking pace,this paper proposes the spatial-temporal probability distribution in the section,and it uses the local probability distribution relation around the target data.The difference effect of global model is minimized,and the structure is light and simple,which is suitable for improving pedestrian recognition effect on the actual intelligent monitoring system.