Objective: This study aimed to enhance the efficiency of indoor evacuation procedures by developing an optimal path-planning model based on the Dijkstra algorithm. The primary goal was to provide a dynamic and intelligent solution for guiding individuals to exit safely and swiftly during emergencies. Considering the complexities of indoor environments and the unpredictability of crowd behavior, the study highlighted the urgent need for a path-planning decision model that can adapt to real-time changes in crowd density and pedestrian flow velocity. Methods: This study employed the single shot detector (SSD) algorithm for person detection and the deep learning-based simple online and real-time tracking (Deep-SORT) algorithm for multi-pedestrian tracking. These approaches, integrated with monocular vision techniques, enabled the extraction of pedestrian counts and flow velocities from architectural surveillance video streams. The Industry Foundation Classes standard was utilized to extract detailed architectural spatial information required for constructing a path network tailored for evacuation purposes. A novel method for predictive edge setting within the existing camera surveillance network was then introduced, allowing for the calculation of residual predictive edge pedestrian flow velocities. This calculation informed the development of an edge weight calculation method for the navigation road network, which is crucial for optimizing evacuation routes. These calculated weights were input into the Dijkstra algorithm to identify the shortest-time evacuation routes from any given node to the nearest exit. Results: The proposed model was validated through experiments conducted using a proprietary pedestrian database under simulated emergency conditions. The SSD algorithm achieved an average precision of 78.2% for pedestrian detection and counting, while the Deep-SORT algorithm achieved multiple object tracking precision and multiple object tracking accuracy scores of 71.2% and 78.5%, respectively. These metrics demonstrate the model's high accuracy in detecting and tracking pedestrians, a crucial aspect of effective evacuation planning. In addition, the model provided optimal path directions at each decision node, allowing for real-time adjustments as conditions change. Moreover, the system could adapt to fluctuating crowd dynamics, a critical feature given the unpredictable nature of human movement during emergencies. Conclusions: This study demonstrated the feasibility of implementing a Dijkstra algorithm-based optimal path-planning model for indoor evacuations, achieving efficient and intelligent route optimization. The research provides a solid theoretical foundation and practical technical support for managing complex indoor evacuation scenarios, thereby contributing to the fields of emergency management and crowd control strategies. Further refinement and application of this model in real-world settings are expected to substantially enhance public safety measures. Additionally, the model's potential integration into existing building management systems presents a promising avenue for improving safety protocols and fostering a more systematic approach to handling indoor evacuations.