WeChat Vehicle Detection Miniprogram

A real-time vehicle detection and matching system built on WeChat Miniprogram with YOLOX + CLIP, powered by WeChat Cloud Development.

Features

• Reference Image Upload - Upload a target vehicle photo and set matching thresholds
• Real-Time Camera Monitoring - Live camera feed with AI-powered vehicle detection (~10s/frame)
• Smart Matching - Composite scoring: image similarity (70%) + color consistency (20%) + brand consistency (10%)
• Detection Reports - View all matches with timestamps, confidence scores, and screenshots
• Task History - Browse and manage past detection tasks
• Local Video Analysis - Batch analyze video files via web interface with higher accuracy model (~25s/frame)
• Scene Enhancement - Automatic image preprocessing for night, fog, rain, and motion blur conditions
• Smart Deduplication - Feature similarity + position tracking + 60s cooldown to avoid duplicate captures

Architecture

+---------------------+     +----------------------+
|  WeChat Miniprogram |---->|  AI Engine (Python)   |
|  Real-time Camera   |     |  YOLOX + CLIP         |
|  Detection Records  |     |  Flask Server :5000   |
+---------------------+     +----------+-----------+
         |                              |
         v                              v
+---------------------+     +----------------------+
|  WeChat Cloud Dev   |     |  Web Video Analysis   |
|  Cloud DB + Storage |     |  localhost:5000/video  |
|  Cloud Functions    |     |                        |
+---------------------+     +----------------------+

Tech Stack

Component	Technology	License
Vehicle Detection	YOLOX-S	Apache 2.0
Feature Matching	CLIP ViT-B/32 + ViT-L/14	MIT
AI Server	Python Flask
Frontend	WeChat Miniprogram (Cloud Dev)
Database	WeChat Cloud Database
Tunnel	natapp

AI Dual-Model Strategy

Scenario	Model	Speed	Accuracy
Miniprogram real-time	ViT-B/32	~10s/frame	Medium
Video file analysis	ViT-L/14	~25s/frame	High
Reference feature extraction	ViT-L/14	One-time	High

Prerequisites

• Python 3.10+
• Node.js
• WeChat DevTools
• natapp (for tunnel forwarding)

Setup

1. Install Python Dependencies

pip install flask torch torchvision Pillow numpy scikit-learn
pip install opencv-python loguru psutil thop tabulate pycocotools
pip install openai-clip

2. Start the AI Engine

cd server/ai_engine
python detection_server.py

On first launch, models will be downloaded automatically:

• YOLOX-S (~80MB)
• CLIP ViT-B/32 (~350MB)
• CLIP ViT-L/14 (~890MB)

3. Configure natapp Tunnel

Download natapp from https://natapp.cn/, set up a tunnel pointing to port 5000, then update the tunnel URL in:

• miniprogram/utils/cloud.js - line 3: AI_ENGINE_URL
• cloudfunctions/uploadReference/index.js - line 9: AI_ENGINE_URL
• cloudfunctions/analyzeFrame/index.js - line 10: AI_ENGINE_URL

4. Configure WeChat Miniprogram

1. Fill in your AppID in project.config.json
2. Fill in your Cloud Environment ID in miniprogram/app.js
3. Open the project in WeChat DevTools
4. Enable Cloud Development and initialize the environment
5. Create database collections: tasks and detections
6. Deploy all cloud functions: right-click each folder under cloudfunctions/ and select "Upload and Deploy: Install Dependencies in Cloud"
7. In local settings, check "Do not verify valid domain names"

5. Database Collection Permissions

In Cloud Development Console > Database > set each collection's permission to:

All users can read and write

Project Structure

├── project.config.json           # WeChat DevTools project config
├── miniprogram/
│   ├── app.js                    # Entry point (cloud init)
│   ├── app.json                  # Page routing + TabBar config
│   ├── app.wxss                  # Global styles
│   ├── utils/
│   │   └── cloud.js              # AI engine communication + cloud DB ops
│   └── pages/
│       ├── index/                # Home: upload reference image + start detection
│       ├── camera/               # Monitoring: camera frames + AI analysis
│       ├── result/               # Results: detection report + sharing
│       └── gallery/              # History: task management
├── cloudfunctions/
│   ├── login/                    # User authentication
│   ├── uploadReference/          # Upload reference image
│   ├── analyzeFrame/             # Analyze single frame
│   ├── getResults/               # Query detection results
│   ├── getRecentTasks/           # Fetch recent tasks
│   └── deleteTask/               # Delete task records
├── server/
│   └── ai_engine/
│       ├── detection_server.py   # AI engine main server
│       └── templates/
│           └── video.html        # Web UI for batch video analysis
└── docs/
    ├── ARCHITECTURE.md           # System architecture details
    └── YOLOX_SETUP.md            # Model setup instructions

Usage

Mode 1: Miniprogram Real-Time Monitoring

1. Open the miniprogram and upload a target vehicle photo
2. Adjust similarity threshold (recommended: 55%)
3. Tap "Start Monitoring" and point the camera at the road
4. The system captures and analyzes one frame every 5 seconds
5. When a match is found: vibration alert + screenshot saved (with green detection box)
6. Tap "Stop" to view the detection report
7. Share the report with WeChat contacts or save to album

Mode 2: Local Video File Analysis

1. Ensure the AI engine is running
2. Open http://localhost:5000/video in your browser
3. Upload a target vehicle photo
4. Enter the video file path (e.g., D:\footage\highway.mp4)
5. Set frame rate and threshold, then click "Start Analysis"
6. Monitor progress, logs, and matched screenshots in real time
7. Open the screenshot folder or download the report when complete

License

• YOLOX: Apache License 2.0
• CLIP: MIT License
• This project: For personal and educational use only. Commercial use is not permitted without prior written consent from the author. If you wish to use this project commercially, please contact the repository owner for authorization.