<p>Lane detection is a vital technology in autonomous driving, crucial for ensuring vehicles remain within designated lanes and reducing accidents. Despite advances, lane detection still faces challenges in complex geometries and extreme conditions. To address these issues, we propose the Lane Detection via Connectivity and Boundary Refinement (CBLane), an enhanced approach that combines connectivity and boundary losses to improve lane boundary detection and continuity. Our method introduces dynamic snake convolution (DSConv) to optimize row and column feature processing, enhancing the recognition of complex lane structures. Additionally, the coordinate attention module is incorporated to strengthen the model’s focus on global and local features. Experimental results demonstrate improvements, with a 0.64% increase in accuracy on the TuSimple dataset, a 1.3% F1-score enhancement on the CULane dataset, and a 1.21% F1-score boost on the CurveLane dataset. The model achieves a speed of 98 FPS, meeting real-time requirements. Here we show that CBLane effectively enhances lane detection robustness and accuracy, paving the way for advanced autonomous driving systems. All source code and datasets will be accessible at <a href="https://github.com/FriendChi/CBLane">https://github.com/FriendChi/CBLane</a>.</p>