Image segmentation is a critical component of computer vision, particularly in autonomous driving scenarios that require precise object identification and spatial understanding. This paper introduces NextSeg, a novel neural network architecture for vision-centric image segmentation that addresses key limitations in existing approaches. Traditional segmentation methods struggle with complex real-world scenes by exhibiting challenges in handling varying lighting conditions, texture-rich environments, and dynamic street scenes. By integrating YOLOv8’s advanced detection paradigms with a modified CondenseNeXt and NextDet architectures, NextSeg offers a more efficient and accurate segmentation solution. The proposed architecture further employs sophisticated techniques including distribution focal loss, Complete-IoU regression, Strip Pooling strategy, and Hierarchical Attentive Feature Aggregation to enhance feature extraction and contextual understanding. Experimental evaluation on the Cityscape dataset demonstrates NextSeg’s superior performance and achieves a mean intersection over union (mIoU) of 79.2% and mean pixel accuracy of 83.7% and outperforms established segmentation models such as FCN, U-Net, PSPnet, and LR-ASPP.