In the context of enhancing autonomous driving, the proactive detection of road obstacles represents a crucial issue for road safety. This study aims to develop an effective method for detecting various mobile obstacles on the road using deep learning. The primary objective of this study is to propose a solution based on Convolutional Neural Networks (CNN) for the detection of road obstacles. CNN-based architecture specializes in the detection of moving obstacles, specifically pedestrians. This approach has been evaluated and compared to existing methods, demonstrating significant improvements under various detection conditions. Our CNN-based architecture significantly outperformed existing detection methods, achieving a precision of 99.85%, a recall of 99.40%, and an F1 score of 99.62. Additionally, we attained an Intersection over Union (IoU) of 85.22% and a mean Average Precision (mAP) of 97.15% at IoU thresholds from 0.5. These results highlight the superior capability of our architectures in identifying and classifying road obstacles, thereby advancing the field of autonomous driving technologies.