Periagoge
Concept
4 min readself knowledge

Computer Vision for Home Inspection: What AI Can (and Can't) Detect

Computer vision AI can analyze property photographs and videos to identify visible defects — water stains, foundation cracks, damaged roofing materials, HVAC issues — that might indicate problems requiring investigation. Understanding what these systems can and cannot detect helps buyers use AI-assisted inspection tools appropriately rather than as replacements for professional physical inspection. This concept covers computer vision inspection as a pre-inspection screening tool with important limitations.

Hypatia
Why It Matters

Computer vision—the subset of AI that processes images and video—has revolutionized property inspection by automating detection of visible defects. Tools can now scan hundreds of photos from a listing and flag potential issues: roof deterioration, foundation cracks, water damage patterns, electrical hazards, and structural concerns. For buyers evaluating properties remotely or screening deals quickly, this technology identifies red flags that might otherwise require an in-person inspection. Understanding what computer vision can reliably detect, and where it fails, prevents both false negatives (missing real problems) and false positives (flagging issues that don't exist).

How Computer Vision Detects Property Defects

Modern computer vision systems use convolutional neural networks (CNNs) trained on hundreds of thousands of labeled images to recognize visual patterns associated with specific defects. A model trained to detect roof damage learns to recognize sagging lines, missing shingles, moss growth, and discoloration patterns that correlate with poor condition. The system breaks images into segments, classifies each segment ("asphalt shingle," "water stain," "rust," "mold"), and assigns confidence scores to its classifications.

The best commercial systems don't just label what they see—they estimate severity. A small water stain might be categorized as "cosmetic," while active mold growth with discoloration spreading across a wall gets classified as "serious." The system can also perform spatial reasoning: it understands that cracks on a foundation's load-bearing section are more concerning than cosmetic cracks on the rim joist.

Where Computer Vision Excels: Visible, Consistent Defects

Computer vision is exceptionally good at detecting defects that create consistent, visually distinct patterns. Roof damage, visible water stains, mold blooms, peeling paint, broken windows, and structural sagging all present distinctive visual signatures that neural networks learn to recognize reliably. If you have quality photos from multiple angles (standard in modern listings), these systems can flag serious issues with 85-95% accuracy for common defects.

The technology shines on volume screening. A buyer evaluating 50 properties can generate AI reports from listing photos in minutes, quickly eliminating properties with obvious red flags without scheduling inspections. This saves enormous time and travel.

The Critical Limitations: Context Blindness and Hidden Defects

Computer vision systems are fundamentally limited by what's visible in photos. They cannot detect:

  • Hidden structural issues: Termite damage inside wall cavities, foundation cracks below grade, pipe corrosion inside walls, electrical code violations hidden behind finished surfaces
  • Deferred maintenance not yet visible: HVAC systems approaching failure, plumbing nearing capacity limits, roof interior rot not yet breaking surface
  • Environmental hazards: Asbestos in insulation, lead paint, radon levels, mold inside wall cavities (only visible mold blooms get detected)
  • Contextual defect severity: A water stain might indicate a historic leak (fixed) or active moisture intrusion (ongoing problem). The image alone doesn't distinguish

Additionally, computer vision struggles with subjective assessments. What counts as "poor condition" vs. "deferred maintenance that's cosmetic"? Different models trained on different datasets will classify the same cracked drywall differently. The systems also frequently misclassify shadows as water damage, or benign patina as rust.

The Photography Bias Problem

AI inspections are only as good as the photos provided. Professional real estate photographers deliberately frame properties to hide problems and highlight strengths. They shoot from angles that minimize unflattering features, use strategic lighting to mask stains, and omit photographs of problem areas entirely. The computer vision system trained on listing photos has inherent selection bias—it's trained on images specifically curated to avoid showing defects.

Conversely, listing photos often miss functional areas entirely (crawlspace, attic access, basement mechanical room) that should be documented for thorough defect detection.

Practical Integration: AI as Screening, Not Verdict

The optimal workflow uses computer vision as a pre-screening layer, not a substitute for professional inspection. Use AI defect detection to identify properties worth investigating further, and to prime your inspector with specific concerns to examine closely. When the AI flags potential foundation cracks, your inspector knows to spend extra time on that section. When the AI finds no obvious red flags, you proceed with more confidence—though still with a professional inspection for properties you're serious about.

For quick portfolio screening (investor evaluating 100+ properties), AI defect detection dramatically compresses timeline. For personal home purchase, where you're making a six-figure decision, use AI reports to inform your inspection scope, not replace it.

Try this: Run 3-5 listing photos from a property you're considering through a computer vision tool (some platforms like Redfin embed this; alternatively, describe the photos to Claude or ChatGPT and ask it to identify potential issues visible in your descriptions). Compile the AI findings, then during your professional home inspection, ask your inspector to specifically examine each flagged area. Note which AI findings were validated by the inspector and which were false positives. This calibrates your understanding of the tool's reliability in your local market and property type.

Helpful guides
Hypatia
Daily Life & Decisions
Related Concepts
Peri
Questions about Computer Vision for Home Inspection: What AI Can (and Can't) Detect?

Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.

Ready to work on Computer Vision for Home Inspection: What AI Can (and Can't) Detect?

Explore related journeys or tell Peri what you're working through.