Research On Pedestrian Re-identification Algorithm Based On Spatiotemporal Information

Person Re-Identification is the use of computer vision technology,in the image or video library retrieve specific pedestrians task,the task has strong practical application demand and significant theoretical research value.At present,person Re-Identification technology mainly relies on appearance information to identify different pedestrians.However,due to the complex and changeable actual scenes,obtaining accurate and robust person Re-Identification is still a very challenging task.The research area of this paper is video-based person Re-Identification,and the main task is to do in-depth research on pedestrian appearance and gait information.The main contributions are as follows:Person Re-Identification aggregated based on apparent features.Compared with images,video sequences contain richer visual information,as well as temporal and spatial information.To make full use of this information,this paper presents a person Re-Identification method based on aggregation of apparent features.This method uses time-domain convolution to extract time information between frames.First,the convolution network is used to extract frame-level features,then the frame-level features are aggregated into sequence-level features by time pooling.At the same time,triple loss and cross-entropy loss are used for joint optimization to learn more discriminant pedestrian features.Experiments on large-scale datasets(MARS,i LDSVID,and PRID-2011)demonstrate the effectiveness of the proposed method.Person Re-Identification based on gait.In order to isolate pedestrian attire,lighting conditions,occlusion,camera imaging and other factors,this paper presents a pedestrian re-identification method based on joint gait spatial-temporal information mining.This method mainly studies space-time information mining and feature space optimization.Start with STM(Spatial-Temporal Module,STM)Re-extracts gait information to compensate for missing information due to occlusion in the frame;then aggregates the frame-level features into sequence-level features using hierarchical convolution network;and finally divides the extracted sequence-level features horizontally using high-level semantic measurement module,which enables the model to focus more on local features,thus extracting more discriminant gait features for feature identification.Great improvement in strength.The method was evaluated on two widely used gait datasets(CASIA-B and OU-LP-Bag).The experimental results demonstrate the validity of this method and significantly improve the gait recognition performance under cross-view and bag walking conditions.Especially on OU-LP-Bag dataset,the accuracy of Rank-1 reaches 93.2%,which is better than the existing technical level.