Archiscan AI: Project Review Document

1. Problem Statement

Manual architectural site surveys are a significant bottleneck in the Architecture, Engineering, and Construction (AEC) industry. The process is traditionally time-consuming, expensive, and susceptible to human error. These inefficiencies lead to project delays, budget overruns, and a high barrier to entry for smaller-scale projects that cannot afford extensive surveying costs.

2. Purpose, Scope, and Objectives

Purpose

Archiscan-AI aims to revolutionize the architectural surveying and planning process by creating an integrated platform that leverages AI and drone technology. Our system automates data capture and generates accurate, regulation-compliant architectural blueprints, streamlining the workflow from site survey to ready-to-use plans.

Scope

The project scope covers the end-to-end process from aerial data acquisition to the final output of 2D CAD drawings and 3D models. The current scope is focused on residential and small-scale commercial properties.

Objectives

Automation: To develop autonomous drone flight paths for efficient and consistent site data collection.
Accuracy: To achieve a high degree of accuracy (error rate < 1%) in the generated blueprints compared to manual methods.
Speed: To drastically reduce the time for a site survey and initial plan generation from weeks to a few days.
Blueprint Generation: To create accurate and detailed 2D CAD floor plans and 3D models directly from the collected drone data.
User Platform: To develop a user-friendly interface for initiating surveys and accessing the generated project data.

3. Abstract (System Overview)

Archiscan-AI presents an innovative solution to the inefficiencies of traditional land surveying. Our system integrates unmanned aerial vehicles (UAVs) with a powerful Artificial Intelligence backend to deliver an automated blueprint generation service. A custom-built drone with a high-resolution camera and RPLIDAR captures comprehensive site data. This data is then processed by a Convolutional Neural Network (CNN), which reconstructs the environment and generates precise 2D floor plans and 3D models. This automated workflow not only saves time and reduces costs but also minimizes the potential for human error.

4. Requirements & Test Environment

Hardware Requirements

Custom-built drone with camera & RPLIDAR A1/A2.
Ground control station (Laptop/PC).
High-performance computing environment (NVIDIA GPU).

Software & Tools

Python, OpenCV, TensorFlow/PyTorch.
DroneKit or similar flight control software.
Datasets: Custom, synthetic, and Matterport3D.

5. System Overview — Proposed System & Outcome

Proposed System Workflow

Input (Data Capture): User defines survey area on our platform, drone flies autonomously capturing images and LiDAR scans.
Process (AI Generation): Data is uploaded to our cloud backend, where CNN algorithms process it to identify structural elements.
Output (Blueprint Delivery): The system generates and delivers regulation-ready 2D CAD files and 3D models to the user's dashboard.

Proposed Outcome

A fully functional platform for commissioning surveys and receiving blueprints.
Blueprint generation time under 48 hours.
High-fidelity 2D plans and 3D models with verifiable accuracy.
Lowered operational costs for builders by up to 60%.

6. System Architecture and Data Flow

The architecture is a modular, cloud-based system consisting of four primary components: the UAV Data Acquisition Unit, the Cloud Processing Platform, the AI Core, and the User-Facing Application.

System Architecture Diagram

Data Flow Diagram (DFD)

Data Flow Steps

User initiates a survey request on the platform.
Platform sends an optimized flight path to the drone.
Drone executes the mission and uploads raw data to cloud storage.
Backend pipeline pre-processes data, runs AI Core analysis, and synthesizes blueprints.
User is notified and can download the final files from the web dashboard.

7. Risk Analysis & Mitigation

Technical Risks:

Risk: Inaccuracy in AI model predictions.
Mitigation: Rigorous training, cross-validation, and human-in-the-loop verification.

Risk: GPS signal loss or sensor malfunction.
Mitigation: Robust RTH safety protocols and pre-flight checks.

Operational Risks:

Risk: Navigating aviation regulations.
Mitigation: Integrate real-time airspace maps for compliance.

Schedule Risks:

Risk: Delays in acquiring training data.
Mitigation: Augment with synthetic data and use transfer learning.

8. Project Plan & Milestones

Phase	Task Description	Timeline
Phase 1: Research	Literature Review, Tech Stack Selection, Drone Design	2 Weeks
Phase 2: Drone Dev	Build and test drone, integrate sensors	3 Weeks
Phase 3: Data Acq.	Collect, pre-process, and label initial datasets	2 Weeks
Phase 4: AI Model	Develop and train the core CNN model	4 Weeks
Phase 5: Platform	Develop backend and user-facing web platform	3 Weeks
Phase 6: Integration	Integrate AI model with the platform and test workflow	2 Weeks
Phase 7: Testing	Conduct real-world beta tests and gather feedback	2 Weeks
Phase 8: Finalize	Bug fixes, documentation, and final presentation prep	1 Week

9. References

Schönberger, J. L., & Frahm, J. M. (2016). Structure-from-motion revisited. In Paper .
Mur-Artal, R., et al. (2015). ORB-SLAM: a versatile and accurate monocular SLAM system. In Paper .
He, K., et al. (2017). Mask R-CNN. In Paper .
A Review on Generative AI in Architecture: The BANI World and Design Education. In Paper .
Using Generative AI for Architectural Design Ideation: A Case Study of Midjourney. In Paper .
Official OpenCV Documentation: docs.opencv.org
Official TensorFlow Documentation: www.tensorflow.org

10. Progress to Date

Phase 1 (Research) - Completed

Conducted literature review and finalized the core technology stack.

Phase 2 (Drone Development) - In Progress

Assembled a custom drone, integrated sensors, and conducted stable flight tests.

Phase 3 (Data Acquisition) - In Progress

Began data acquisition flights and are pre-processing/labeling the initial dataset.