Abstract. Addressing the challenge of quantitatively analyzing and presenting pedestrian elements within open community spaces is of significant importance. Focusing on the indoor scene spaces of open communities, this study introduces the TJ-Person pedestrian target recognition image dataset. Furthermore, we design a deep learning-based community pedestrian activity analysis network model and incorporate various attention mechanisms, such as SA, CA, CBAM, SE, and SK, into the YOLO v5s deep learning target recognition network framework for comparative evaluation of pedestrian target recognition in open communities. Utilizing the optimized YOLO Swin Transformer Person (YOLO-STP) network, precise identification of pedestrian targets across multiple scenarios was achieved. We conducted experimental verification using four typical scenarios within Shanghai's NICE2035 open community as case studies. The results demonstrated that the proposed YOLO-STP community pedestrian activity analysis network model achieved an optimal detection accuracy of up to 98.47%. In all four tested scenarios, the YOLO-STP method consistently exhibited competitive performance. Moreover, in the COCO-2017 open-source dataset testing, the YOLO-STP method outperformed other networks of the same type, showcasing its significant advantages. Overall, the research presented in this study provides a crucial technical foundation for the analysis and recognition of pedestrian targets in future community scenarios.