Abstract:Real-time and high-performance segmentation is a crucial but challenging perception task for computationally constrained robots, such as the humanoid NAO robot used in the RoboCup Soccer Standard Platform League. However, most existing convolutional neural network (CNN)-based models for semantic segmentation suffer from massive computational costs, which prevents them from being applied to performing real-time inference with a NAO. In this article, we first publish meticulously annotated datasets for training and evaluating semantic segmentation models. Then, we propose a fast downsampling module that downsamples the image while maintaining the spatial information and a novel dense learning module that learns high-level semantic information while recovering the spatial details. Based on these operations, by using a multiscale fusion method to recover the resolution, we propose a more efficient and real-time segmentation model called RoboSeg primarily aimed at offering better speed and accuracy tradeoffs. Finally, to accommodate practical engineering applications, we offer a promising deployment guideline for the CNN model describing how to deploy it on computational resource-limited robots and achieve real-time performance. The experimental results show that the RoboSeg exceeds the state-of-the-art networks in RoboCup scene segmentation: we attain a mean IoU of 87.35% and a pixel accuracy of 96.88% on our dataset using a model that contains only 0.29M parameters and performs just 0.73 GFLOPs. Under the proposed deployment strategies, the network can run at above 30 FPS on NAO robots with downsampled frames.

Robust Color Segmentation Through Adaptive Color Distribution Transformation

A Global Vision System: Using Hue Thresholds to Exact Feature and Recognize

Multi-objects Real Time Recognition Based on Color Information

A Training Method for Robust Recognition of Color Uniforms of Soccer Robots

Fast Recognition and Precise Localization of Humanoid Soccer Robot Vision System

Robustifying Semantic Cognition of Traversability Across Wearable RGB-depth Cameras

An Autonomous Robot Oriented Real-Time Color Vision System

Robust image segmentation against complex color distribution

A robust image segmentation method using hierarchical color clustering.

A Monocular Vision System for Playing Soccer in Low Color Information Environments

Autonomous Recognition and Positioning of Shield Segments Based on Red, Green, Blue and Depth Information

Research in Target Recognition Arithmetic of Robot Based on Color

Fuzzy color recognition and segmentation of robot vision scene

Color learning and illumination invariance on mobile robots: A survey

Robust Visual Localization of a Humanoid Robot in a Symmetric Space

RGB-D Based Object Segmentation in Severe Color Degraded Environment.

Shape and Color Object Tracking for Real-Time Robotic Navigation

SG-Grasp: Semantic Segmentation Guided Robotic Grasp Oriented to Weakly Textured Objects Based on Visual Perception Sensors

Color Recognition for Rubik's Cube Robot

Confidence-Based Color Modeling for Online Video Segmentation

RoboSeg: Real-Time Semantic Segmentation on Computationally Constrained Robots