Significant progress has been made in monocular 3D detection technology, but it still faces challenges. One key challenge is accurately estimating the 3D coordinates and velocity of objects due to the lack of depth and motion information from monocular cameras. Additionally, monocular 3D detection suffers from lower accuracy caused by positioning errors, scale uncertainty, and occlusion. In this study, we propose a lightweight and enhanced monocular FCOS3D detection algorithm, called MultiBranch-Moni3D. The algorithm incorporates improvements such as the integration of the SE module EfficientNet and the replacement of the FPN structure with BiFPN for improved feature representation. Furthermore, we modify the prediction approach by adopting direct prediction instead of binary prediction, reducing the need for positive sample determination. Experimental evaluations on the NuScenes dataset demonstrate promising results. Our proposed algorithm achieves a notable reduction in inference time from 79ms to 27ms on a single NVIDIA RTX3090 video card. Although there is a slight decrease in mAP by 1.71%, other metrics, such as mATE, mASE, mAAE, mAVE, and mAOE, experience increases of 2.57%, 4.4%, 6.72%, 11.1 %, and 27.5%, respectively. Overall, our proposed enhancements aim to enhance the efficiency and accuracy of monocular 3D detection. While there is a minor trade-off in detection accuracy, the algorithm's improved processing speed makes it a promising approach for real-world applications in the field of monocular 3D detection.