Research Article
Pedestrian Detection in Surveillance Video Based on Time Series Model
@INPROCEEDINGS{10.1007/978-3-031-50574-4_8, author={Hui Liu and Liyi Xie}, title={Pedestrian Detection in Surveillance Video Based on Time Series Model}, proceedings={Multimedia Technology and Enhanced Learning. 5th EAI International Conference, ICMTEL 2023, Leicester, UK, April 28-29, 2023, Proceedings, Part II}, proceedings_a={ICMTEL PART 2}, year={2024}, month={2}, keywords={Time Series Model Monitoring Video Pedestrian Detection Region of Interest}, doi={10.1007/978-3-031-50574-4_8} }- Hui Liu
Liyi Xie
Year: 2024
Pedestrian Detection in Surveillance Video Based on Time Series Model
ICMTEL PART 2
Springer
DOI: 10.1007/978-3-031-50574-4_8
Abstract
To solve the problem of low detection accuracy caused by the occlusion of downlink people in complex scenes, a pedestrian detection method based on time series model in surveillance video is proposed. The gray values of pixels at the same location are regarded as a time series, and the mixed Gaussian model is used to recognize the pedestrian foreground. The threshold segmentation method is used to segment the image. The threshold segmented image is projected vertically to the X axis, and the overlapped pedestrian trough is used as the segmentation point. The bimodal feature window is projected vertically to segment the region of interest. Mark the feature points of the real image dataset, build an enhanced feature point detection network model, and obtain the descriptor detection results. The time-domain and frequency-domain information is represented as a symbol sequence, and the target is clustered using the equal length overlapping time window segmentation method, so that the location center of gravity does not change and the specified convergence degree is achieved. Balance the data fusion features to determine the pedestrian detection results in the surveillance video. The experimental results show that this method can detect all pedestrians, the maximum accumulated error of target recognition is 21%, and the maximum average accuracy of target matching is 90%, which proves that the detection effect is good.
