Unauthorized access to the bidding evaluation office and the departure of experts can significantly compromise the quality of on-site assessment work and introduce substantial integrity risks. To address these concerns, this paper presents the development of a hybrid neural network, integrating Yolov5s, Deepsort, and Dlib to ascertain the status of person within the bidding evaluation office. Our approach is bifurcated into two primary components. Firstly, Deepsort is integrated with Yolov5s to develop a model for the detection and tracking of personnel within the evaluation office. The model detects, counts, and tracks the flow of personnel on site and assesses the presence or absence of experts. Subsequently, Yolov5s, enhanced by the Swin Transformer architecture, refines the Dlib facial recognition model, augmenting its capacity to detect and swiftly identify small faces, thereby discerning potential intruders. The experimental results demonstrate that the model is capable of effectively detecting and tracking personnel on site, recognizing novel micro-scale targets, verifying individual identities, and evaluating the status of on-site personnel. Throughout the testing phase, when juxtaposed with conventional methodologies, our model has exhibited a marked enhancement in the accuracy of detecting unauthorized entry and the absence of designated experts, enabling real-time analysis of the evaluation office's status.