Due to poor lighting conditions at night, complex scenes, potential occlusion of human forms, and the tendency of moving objects to produce ghosting effects, video surveillance often faces challenges such as suboptimal detection rates, detection errors, and missed detections in the tasks of identifying and detecting nighttime pedestrian falls. To address these issues, the original YOLOv8(you only look once version 8) base model, the cross-scale feature fusion module (CCFM) was introduced into the Neck section of YOLOv8, resulting in a new detection model called CCFM-YOLOv8n. To enhance the model's ability to detect dynamic feature scales and small targets while maintaining high detection speed. Additionally, the global attention mechanism (GAM) was incorporated into the Backbone section of the model to improve the overall performance of the deep neural network. Finally, the improved model introduced the minimum point distance intersection over union (MPDIoU) loss function to enhance the bounding box regression capability of the network model. Experimental results demonstrate that the improved algorithm achieves a detection accuracy of 96.1% for nighttime fall targets, with the recall rate increasing to 92.4%. Compared to YOLOv8n, the mAP@0.5:0.95 metric improved by 3.8 percentage points, while the parameter count decreased by 19.7 percentage points. This fully demonstrates that the improved model offers greater advantages over the original model in the practical application of video surveillance for nighttime pedestrian fall detection.