Pedestrian safety in urban areas poses a considerable challenge due to the prevalence of traffic congestion and the associated risks of accidents. This study aims to enhance pedestrian detection (PD) by integrating Explainable Artificial Intelligence (XAI) with advanced deep learning (DL) techniques. By utilizing the Penn-Fudan Dataset, we optimized three convolutional neural network (CNN) architectures, ResNet50, DenseNet121, and VGG16, to effectively recognize pedestrians in complex urban landscapes. To ensure robustness, we validated the models using the Caltech Pedestrian Dataset, which provides diverse urban pedestrian scenarios. We applied XAI methodologies, including Gradient-weighted Class Activation Mapping (Grad-CAM) and Local Interpretable Model-agnostic Explanations (LIME), to clarify the decision-making processes of the models and to enhance transparency in PD. The results demonstrate that ResNet50 outperforms the other architectures in accuracy, while the XAI approaches provide essential insights into the PD process. This research underscores the potential of XAI to improve model transparency and facilitate real-time decision-making in PD systems.