Automated crack detection in roads and infrastructure is crucial for ensuring safety and longevity in civil engineering. This study introduces LightCrackNet, a novel lightweight deep learning model designed for real-time crack detection using embedded devices. Our approach addresses the need for efficient, on-site monitoring by enabling rapid processing and accurate identification of cracks within the constraints of embedded systems. We employ a multi-scale feature extraction strategy that captures both fine details and broader contextual information, facilitating precise segmentation under diverse conditions. Trained and evaluated on over 10 publicly available datasets, LightCrackNet produces highly competitive results among existing crack detection models, particularly in resource-constrained environments. Extensive testing demonstrates that our model achieves high accuracy while maintaining low computational overhead, making it ideal for deployment on mobile inspection platforms such as drones and handheld devices. This research enhances the capabilities of automated infrastructure monitoring and promotes proactive maintenance practices, ultimately contributing to safer transportation networks and structures.