DAIS: Deep Learning-Based Detection of Dog-Human-Vehicle Interactions in Urban Surveillance

Stray dogs pose increasing public safety concerns in urban environments, often engaging in aggressive behaviors such as chasing pedestrians and vehicles. Traditional animal control approaches are insufficient to handle this growing problem, necessitating intelligent and automated surveillance solutions. This paper introduces a novel deep learning-based framework capable of real-time detection and analysis of interactions among dogs, pedestrians, and vehicles. The proposed system integrates a fine-tuned YOLO-based object detector for accurate recognition of relevant entities, a CNN-based classifier for dog breed identification, and the DeepSORT tracking algorithm enhanced by Kalman filtering for robust multi-object tracking. Additionally, a novel target interaction association algorithm isolates relevant object pairs, while an LSTM-based temporal model classifies interaction sequences to infer aggressive or pursuit behaviors. Experimental evaluations confirm the effectiveness and reliability of the proposed framework, highlighting its potential to significantly improve public safety in urban areas.