<![CDATA[Augmented reality (AR) indoor navigation provides visual guidance, while crowd-responsive route selection requires current crowd information to be connected to route planning. This study evaluates the integration of camera-based crowd monitoring with AR route adaptation and measures its response time on local infrastructure. The proposed system uses YOLOv8s to detect people, classifies crowd levels from person counts, and uses Unity NavMesh to compute alternative routes displayed through AR on a smartphone. The integration extends static routing by using crowd levels estimated from camera feeds to temporarily exclude highly crowded passages from route planning and update the displayed AR route when an alternative path is available. Evaluation covered detection, crowd-level classification, rerouting, response time, and a static–dynamic routing comparison. Detection mAP@0.5 was 91.92%, and crowd-level classification accuracy was 90.67% across 75 observations. Rerouting succeeded in all 30 controlled events, with an average response time of 2.772 s from the first classification of a zone as highly crowded to the display of the updated AR route. The suitability of this response time for hospital navigation requires further evaluation. In 30 paired trials, dynamic routing avoided the designated crowded zones but increased mean path length and navigation time in all three zones. Testing was limited to one hospital floor with three monitored zones. User comfort, accessibility, and performance in larger buildings require further evaluation.]]>
