Semantic Vector Sequence Localization

中文 | English

Contents

• Overview
• Visual overview
• Example results
• Repository structure
• Related publication
• Citation
• License
• 中文说明

English

A MATLAB / C++ research prototype for visual place recognition and localization using semantic vector representations and sequence-level matching.

The core idea is to represent a driving scene with compact semantic descriptors, then use temporal sequence information to reduce ambiguity during localization.

Motivation

Single-frame place recognition can be ambiguous in road environments: many streets, lanes, buildings, or parking-lot scenes can look similar. Sequence-level localization uses temporal context to make matching more robust.

Key ideas

• Build semantic-vector descriptors from visual observations.
• Match query sequences against reference sequences.
• Use temporal consistency for more reliable place recognition.
• Evaluate localization results with analysis scripts.

Visual overview

Example results

The repository contains result plots comparing different semantic-vector and sequence-matching configurations for place-recognition / localization experiments.

Repository structure

.
├── Analyse/                  # Result analysis utilities
├── PreDataOfPark/            # Data preprocessing utilities
├── datasets/                 # Small historical experiment dataset
└── README.md

Keywords

visual localization, place recognition, sequence localization, semantic vector, autonomous driving, robotics navigation, MATLAB, C++

Related publication

This repository contains experimental code related to the ideas in the following paper:

• Visual Place Recognition via Semantic and Geometric Descriptor for Automated Valet Parking

The code should be treated as a research prototype / partial implementation, not as an official or complete reproduction of the paper.

Getting started

The original experiments were run in a MATLAB + C++ research environment. To adapt the project, start from the preprocessing scripts under PreDataOfPark/, then inspect the analysis utilities under Analyse/ and update local dataset paths.

Citation

If you use this repository, please cite or acknowledge it using the metadata in CITATION.cff.

License

This repository is released under the Apache License 2.0. Please retain the license and notice files when redistributing or reusing the code.

---

中文说明

这是一个基于 MATLAB / C++ 的视觉定位研究原型，使用 语义向量表示 和 序列匹配 来做地点识别与定位。

项目的核心想法是：道路场景中的单帧图像可能存在较强歧义，而连续帧序列可以提供时间一致性，从而提升定位和地点识别的鲁棒性。

关键词

视觉定位、地点识别、序列定位、语义向量、自动驾驶、机器人导航、MATLAB、C++。

相关论文

该仓库包含与以下论文思想相关的实验代码：

• Visual Place Recognition via Semantic and Geometric Descriptor for Automated Valet Parking

需要注意：该仓库更适合作为研究原型 / 部分实现来看待，不应理解为论文的官方完整复现。

引用与许可

如果你使用该仓库，请通过 CITATION.cff 引用或致谢该项目。许可协议见 LICENSE。