Low-altitude drones, which are unimpeded by traffic congestion or urban terrain, have become a critical asset in emergency rescue missions. To address the current lack of emergency rescue data, UAV aerial videos were collected to create an experimental dataset for action classification and localization annotation. A total of 5082 keyframes were labeled with 1–5 targets each, and 14,412 instances of data were prepared (including flight altitude and camera angles) for action classification and position annotation. To mitigate the challenges posed by high-resolution drone footage with excessive redundant information, we propose the SlowFast-Traffic (SF-T) framework, a spatio-temporal sequence-based algorithm for recognizing traffic accident rescue actions. For more efficient extraction of target–background correlation features, we introduce the Actor-Centric Relation Network (ACRN) module, which employs temporal max pooling to enhance the time-dimensional features of static backgrounds, significantly reducing redundancy-induced interference. Additionally, smaller ROI feature map outputs are adopted to boost computational speed. To tackle class imbalance in incident samples, we integrate a Class-Balanced Focal Loss (CB-Focal Loss) function, effectively resolving rare-action recognition in specific rescue scenarios. We replace the original Faster R-CNN with YOLOX-s to improve the target detection rate. On our proposed dataset, the SF-T model achieves a mean average precision (mAP) of 83.9%, which is 8.5% higher than that of the standard SlowFast architecture while maintaining a processing speed of 34.9 tasks/s. Both accuracy-related metrics and computational efficiency are substantially improved. The proposed method demonstrates strong robustness and real-time analysis capabilities for modern traffic rescue action recognition.

In the domain of emergency rescue, the capacity to swiftly and precisely identify the actions of rescue personnel is paramount to optimising the efficacy of rescue operations. This study proposes an enhanced real-time action recognition algorithm, termed DFAF-YOLO (Dynamic Feature Extraction and Multiscale Attention Fusion YOLO), in response to the shortcomings of existing algorithms in complex scenarios, particularly their inadequate recognition accuracy. The proposed algorithm incorporates several core optimisations with a focus on feature fusion. Firstly, the CBAM attention module is introduced in the algorithm’s Neck region, with the objective of suppressing redundant information and enhancing core feature transmission through the dual effect of channel and spatial attention. Secondly, the ASFF mechanism is introduced in the Head detection region, with the aim of enabling dynamic alignment and weight adjustment of multi-scale feature maps transmitted from the Neck. This mechanism addresses the issue of information loss caused by scale mismatch. The two modules work in synergy to enhance key feature extraction, with the WIoU loss function being introduced to optimise the accuracy of detection box regression. The experimental validation process has demonstrated that the proposed scheme attains an mAP_0.5 of 88.6% and an accuracy of 86.5% on the self-built rescue action dataset. In comparison with the base YOLOv8n model, the proposed model demonstrates a 4.4% enhancement in mAP_0.5 and a 2.9% enhancement in detection accuracy,verifying the effectiveness of the designed optimisations in improving recognition performance under complex rescue scenarios.