Abstract: Monocular depth estimation is a core technology for 3D environmental perception and is of significant value in applications such as autonomous driving and robot navigation. As a representative lightweight model in this field, Lite-Mono achieved an excellent balance between performance and efficiency through use of dilated convolutions and a local-global attention mechanism. However, two major bottlenecks in current Lite-Mono architecture were identified. Firstly, fixed dilation rate in its dilated convolutions led to mismatch between receptive field and target scales. Small objects lost detailed features due to excessively large rate, while large objects lacked contextual information due to insufficient receptive field. Secondly, high computational complexity of O(H2W2C) in local-global attention module hindered processing of high-resolution inputs, thus limiting real-time application. To address these issues, an enhanced model, DDHL (Dynamic Dilated convolution and Hierarchical Local-global attention) model based on Lite-Mono was proposed. It introduces a Dynamic Dilated Convolution module, which first adjusts the receptive field in real-time via a fusion weight rate predictor and generates adaptive weights by combining channel attention. Furthermore, a Hierarchical Local-Global Attention module was designed to reduce computational complexity to O((HW/M)^2C). Additionally, a learnable global token was incorporated to establish cross-window dependencies. Experimental results on Make3D dataset demonstrate that DDHL model achieves significant improvements on key metrics. Absolute Relative Error (Abs Rel) is reduced from 0.462 to 0.290, a decrease of 37.2%. Squared Relative Error (Sq Rel) is decreased by 49.7%. Improved model achieves better balance between accuracy and efficiency, demonstrating considerable practical application value.