Answers are grounded in this paper and cite their sources.
Demonstrates improved semantic segmentation efficiency in mobile robots, suggesting advancements in navigation capabilities.
One of the key challenges in semantic segmentation for path planning of mobile robots (MR) lies in deploying effective segmentation models on embedded automatic systems with limited processing capability. This study suitably improves a Binary Neural Network (BNN)-based semantic segmentation for monocular images, with the proposed backbone being ResNet18 and incorporating an efficient decoder based on PSP-Net. In detail, by reducing processing parameter magnitudes and optimizing network propagation, the proposed method enhances training efficiency, speeds up inference, and reduces computational cost. The introduced method’s outcomes outperform those of convolutional neural network (CNN) architectures, particularly with a notable 3.11-fold increase in inference velocity while maintaining an accuracy of 97.7% and mean intersection over union (mIoU) of 74.36% based on Cityscapes dataset. Moreover, BNNs facilitate the deployment of semantic segmentation models on hardware with restricted physical or resource capacities. The proposed method achieves an intersection over union (IoU) and Dice scores of 83.14% and 88.21%, respectively, with throughput improvements of 1.6 to 2.5 times and latency reduced to 0.2 times the average of other models evaluated. From experimental results, the suggested Binary-SegNet framework can be well integrated in intelligent MR, especially in advancing knowledge systems for effective MR’s navigation in diverse environments.
No takes yet. Share an insight, caveat, or question.
Dang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: