Image Capture and Analysis

This repository provides a modular Python workflow for street view image capture, LLM-based scene analysis, and coordinate refinement, designed to support precision mapping of urban curb ramps and accessibility barriers.
The workflow replicates the manual logic of field observation through multi-round sampling and reasoning, ensuring complete and accurate coverage of complex intersections.

Overview

The system combines API-driven street view acquisition, large language model (LLM) scene understanding, and coordinate offset algorithms to identify and analyze accessibility-related urban features such as ramps and curbs.

Core Workflow

1. Initial Image Capture
   Retrieve first-round street view images via Baidu Maps API.
2. Scene Parsing with LLM
   Use a language model (e.g., Doubao LLM) to interpret the visual context of street scenes, identifying potential curbs, ramps, and missing accessibility elements.
3. Adaptive Re-sampling
   Apply geometric and semantic reasoning to compute second- and third-round sampling points, ensuring comprehensive visual coverage through coordinate offsets.
4. Automated Data Integration
   Merge multi-round outputs with intersection metadata for final analysis and model training.

---

🧩 Repository Structure

📂 image-capture-and-analysis/
├── 1_baiduStreetViewSpider_simple.py # Basic Baidu Street View crawler (first-round sampling)
├── 2_读取照片文件名称.py # Reads and manages downloaded image filenames
├── 3_find_same_center.py # Detects overlapping or duplicate image centers
├── 4_crawl_and_analyze_doubao_LLM.py # Integrates Doubao LLM for scene interpretation and parsing
├── 5_合并一轮输出和路口信息.py # Merges first-round outputs with intersection metadata
├── 6_采样点偏移（二轮爬取准备）.py # Applies coordinate offset for second-round resampling
├── 7_baiduStreetViewSpider_需读取爬取角度.py # Extended crawler supporting specified camera yaw/pitch
├── 8_对面路口上下文推理.py # Cross-intersection contextual reasoning module
├── 9_第三轮关联处理.py # Third-round linkage and post-processing script
└── README.md

---

Requirements

• Python 3.8+
• Recommended packages:

  requests
  pandas
  numpy
  openai (for LLM API)
  opencv-python
  tqdm

• Optional: Doubao LLM API access; Baidu Maps API key