We present an integrated multi-modal safety assessment framework for autonomous mobile robots navigating complex environments. Our system combines traffic sign recognition, dynamic object detection, and semantic environment segmentation to enable a thorough understanding of the environment, thus facilitating safety assessments. By integrating multiple perception modalities, we achieve a more robust approach to safe navigation that outperforms single-module systems. A custom-built Convolutional Neural Network is trained on an augmented dataset that merges the German Traffic Sign Recognition Benchmark and newly collected Albanian traffic sign images, achieving state-of-the-art performance with reduced computational complexity. To guide navigation in complex terrains with dynamic objects, we implement a YOLO-based algorithm optimized for pedestrian and vehicle detection. A DeepLabv3+ based semantic segmentation component classifies the scene into contextual areas relevant for safe navigation. Outputs from individual modules are integrated using a network-based safety classifier to identify scenes as “safe” or “unsafe” for navigation. The multi-modal system leads to improved navigation safety assessment, accurate object classification and detection” and context-aware navigation.
Safety Analysis for Autonomous Mobile Robot Navigation Using Traffic Sign Recognition, Dynamic Object Detection, and Semantic Segmentation
2024-12-18
4209715 byte
Conference paper
Electronic Resource
English
Semantic Segmentation to Develop an Indoor Navigation System for an Autonomous Mobile Robot
BASE | 2020
|Semantic Segmentation to Develop an Indoor Navigation System for an Autonomous Mobile Robot
BASE | 2020
|Traffic Sign Detection and Recognition Using Ensemble Object Detection Models
Springer Verlag | 2023
|Sign detection for autonomous navigation
SPIE | 2003
|