Ensuring pedestrian safety is vital in intelligent transportation, especially with the growing adoption of advanced driver assistance systems. This paper proposes a novel vision-based framework that integrates a real-time object detection model, You Only Look Once version 8, with a sequential learning model known as Long Short-Term Memory networks to analyze motion patterns and predict pedestrian crossing behavior. The system combines spatial and temporal features extracted from video data, enabling context-aware intention prediction. Experimental evaluations using a publicly available pedestrian behavior dataset confirm the framework’s effectiveness in identifying dynamic pedestrian actions across diverse traffic scenes. This method demonstrates strong potential for deployment in autonomous vehicles and intelligent traffic systems to enhance real-time decision-making and improve pedestrian safety.