Performance Evaluation of DeeplabV3 Plus for Road Segmentation in iCar ITS

Road segmentation plays a vital role in enabling navigation of Intelligent Car ITS (iCar ITS), especially in complex and low-light environments. However, many previous studies have focused solely on accuracy metrics while overlooking real-time inference performance on low-end hardware and the lack of road datasets from Southeast Asian regions. Although lightweight models such as MobileNetV2 have been previously explored, evaluations of newer architectures like MobileNetV3-Large remain limited in constrained resource scenarios. This study investigates the performance of DeepLabV3 Plus with three backbone networks ResNet-50, ResNet-101, and MobileNetV3-Large trained using supervised learning on a custom dataset collected via the iCar ITS platform in Indonesian urban settings, particularly at night to simulate challenging conditions. The training was conducted using the Adam optimizer in Google Colab, and inference was evaluated on a laptop (Intel Core i7-8550U, 22.2 GB RAM, NVIDIA GeForce MX130) to assess real world feasibility. Experimental results show that ResNet-50 achieved the highest segmentation accuracy (Pixel Accuracy: 99.64%, mIoU: 76.45%, Mean F1: 82.03%), while MobileNetV3-Large maintained competitive accuracy (PA: 99.59%, mIoU: 74.96%, Mean F1: 80.28%) with significantly better real-time performance (FPS: 29.13, processing time: 0.0187 s), making it more suitable for embedded applications. These findings indicate a trade-off between accuracy and computational efficiency; this is important because, in addition to high accuracy, real-time performance and stable road detection are essential to support the system's response to fast-moving vehicles and dynamic road conditions.

• Publication year

  2025

• Venue

  International Seminar on Intelligent Technology and Its Applications

• Publication date

  2025-07-23

• Fields of study

  Not labeled

• Identifiers
  DOI 10.1109/ISITIA66279.2025.11137569
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Sensor Fusion System for Localization of Autonomous Car

  2024cited by this paper
• Enhancing Road Scene Segmentation With an Optimized DeepLabV3+

  2024influential reference
• Semantic Segmentation of Road Traffic Sign based on Improved Deeplabv3+

  2024cited by this paper
• Road Image Segmentation using Unmanned Aerial Vehicle Images and DeepLab V3+ Semantic Segmentation Model

  2021cited by this paper
• Artificial intelligence applications in the development of autonomous vehicles: a survey

  2020cited by this paper
• Semantic scene segmentation in unstructured environment with modified DeepLabV3+

  2020influential reference
• Identification and enumeration of cyanobacteria species using a deep neural network

  2020cited by this paper
• Road Segmentation Based on Hybrid Convolutional Network for High-Resolution Visible Remote Sensing Image

  2019cited by this paper
• Autonomous Cars: Research Results, Issues, and Future Challenges

  2019cited by this paper
• Road Segmentation of Unmanned Aerial Vehicle Remote Sensing Images Using Adversarial Network With Multiscale Context Aggregation

  2019cited by this paper
• Semantic Segmentation of UAV Aerial Videos using Convolutional Neural Networks

  2019cited by this paper
• Deep neural network ensemble for pneumonia localization from a large-scale chest x-ray database

  2019cited by this paper
• The Effectiveness of Data Augmentation in Image Classification using Deep Learning

  2017cited by this paper
• Rethinking Atrous Convolution for Semantic Image Segmentation

  2017cited by this paper
• Learning Aerial Image Segmentation From Online Maps

  2017cited by this paper
• Is the deconvolution layer the same as a convolutional layer?

  2016cited by this paper
• HD Maps: Fine-Grained Road Segmentation by Parsing Ground and Aerial Images

  2016cited by this paper
• Multi-Scale Context Aggregation by Dilated Convolutions

  2015cited by this paper
• Beyond pixels: A comprehensive survey from bottom-up to semantic image segmentation and cosegmentation

  2015cited by this paper
• Learning Deconvolution Network for Semantic Segmentation

  2015cited by this paper
• Path Following for a Car-like Mobile Robot based onFuzzy-logic

  2012cited by this paper

• No citing papers are available for this paper.