This paper proposes a single-stage text detection method for low-light environments, which avoids the need for low-light image enhancement. The method introduces a spatial-constrained learning module during training to preserve spatial information of text, ensuring the text detector can accurately identify text's local topological features. The approach incorporates dynamic snake feature pyramid networks and a bottom-up contour shaping strategy with a novel rectangular accumulation technique for accurate delineation of text features. Additionally, a comprehensive low-light dataset for arbitrary-shaped text, LATeD, is introduced, containing 13,923 multilingual and arbitrary shape texts across diverse low-light scenes. The method achieves state-of-the-art results on this dataset and performs comparably on standard normal light datasets. The code and dataset are released. The paper also compares the proposed method with existing approaches and demonstrates its effectiveness in both low-light and normal-light text detection. The method's success is attributed to its ability to capture text's topological and streamline features through the integration of spatial-constrained modeling, dynamic snake feature pyramid networks, and bottom-up contour shaping. The results show that the method is robust in detecting curved texts in low-light environments and outperforms existing methods in both low-light and normal-light text detection.This paper proposes a single-stage text detection method for low-light environments, which avoids the need for low-light image enhancement. The method introduces a spatial-constrained learning module during training to preserve spatial information of text, ensuring the text detector can accurately identify text's local topological features. The approach incorporates dynamic snake feature pyramid networks and a bottom-up contour shaping strategy with a novel rectangular accumulation technique for accurate delineation of text features. Additionally, a comprehensive low-light dataset for arbitrary-shaped text, LATeD, is introduced, containing 13,923 multilingual and arbitrary shape texts across diverse low-light scenes. The method achieves state-of-the-art results on this dataset and performs comparably on standard normal light datasets. The code and dataset are released. The paper also compares the proposed method with existing approaches and demonstrates its effectiveness in both low-light and normal-light text detection. The method's success is attributed to its ability to capture text's topological and streamline features through the integration of spatial-constrained modeling, dynamic snake feature pyramid networks, and bottom-up contour shaping. The results show that the method is robust in detecting curved texts in low-light environments and outperforms existing methods in both low-light and normal-light text detection.