Make computer vision explainable
Students inspect pixels, masks, contours, position, motion, markers, and confidence evidence instead of treating a camera result as magic. The system must distinguish a target, no target, and uncertainty.
ROBOTICS TRACK / LEVEL 05 ROADMAP
Computer Vision Lab
Treat camera images as measurable data, build an inspectable processing pipeline, make useful decisions, and report uncertainty honestly.
Roadmap status: camera platform validation pending
This is a standalone computer-vision plan, not a released course or hardware recommendation. Both the primary Raspberry Pi path and a suitable computer/webcam fallback must prove acquisition, restart, offline use, recovery, and the complete vision baseline. Do not purchase Level 5 hardware from this page yet.
Core programming and evidence skills transfer
- Meet Level 4 academic readiness or show equivalent conditions, loops, functions, states, calibration, logging, and evidence.
- Work with row/column grids, indexes, comparisons, and ranges.
- Complete the required Python Bridge before Lesson 1.
- Follow family camera-privacy rules and distinguish local evidence from public sharing.
REQUIRED BRIDGE
Python before the camera
Translate familiar ideas into Python
Run and stop a script, change and print a variable, use indentation and if/else, loop over a list, define a function, import a verified module, read a short traceback, and restore a known-good file.
Not a thirteenth vision lesson
The bridge uses ordinary .py files without a camera. It creates a dependable language and recovery baseline so Lesson 1 can focus on camera acquisition instead of setup confusion.
PROVISIONAL HARDWARE
Two possible offline paths
Raspberry Pi plus Camera Module 3
The preferred dedicated setup is planned around a Raspberry Pi 4-class computer and standard Camera Module 3, with a known-good local display, storage, power, and recovery workflow.
A suitable computer and UVC webcam
A family may be able to reuse an existing computer and compatible webcam if that path passes the same offline acquisition, module, restart, evidence, and privacy requirements.
Planning estimate: $160–$235 incremental for the dedicated primary path, or $0–$60 when a suitable computer and webcam can be reused. Neither path is approved for purchasing yet.
PLANNED LEARNING ARC
Twelve lessons
Lesson 01 / CameraAcquire, display, save, close, and restart
Use the parent-preflighted path and prove a clean recovery after restart.
Lesson 02 / ArraysPixels, dimensions, coordinates, and regions
Connect an image to NumPy-array shape and row/column indexing.
Lesson 03 / LightChannels, brightness, and histograms
Measure how lighting changes the data before selecting a rule.
Lesson 04 / MaskHSV thresholding and Boolean masks
Turn a stated color range into an inspectable yes/no image.
Lesson 05 / ObjectsContours, area, and bounding boxes
Filter false detections using measured size and shape evidence.
Lesson 06 / PositionCentroid and position error
Measure where a target is relative to the image center.
Lesson 07 / MotionFrame difference
Compare images over time and separate useful motion from ordinary variation.
Lesson 08 / MarkersQR or visual-marker reading and validation
Accept only recognized, checked information rather than any apparent pattern.
Lesson 09 / UncertaintyConfidence proxies and explicit UNSURE
Define when evidence is too weak to support a confident result.
Lesson 10 / EvidenceAnnotated images and timestamped logs
Save useful evidence at a bounded rate without flooding storage.
Lesson 11 / IntegrateSmart-camera prototype
Assemble acquisition, processing, decision, uncertainty, display, and evidence stages.
Lesson 12 / ImproveUser test, revision, privacy, and limitations
Test in more than one environment and explain false positives, false negatives, and limits.
CAPSTONE + READINESS
Smart camera
Detect, explain, and admit uncertainty
The planned smart camera detects or tracks chosen evidence, creates a rate-limited annotated image and timestamped log, displays what it believes, and reports when confidence is too low.
Defend the processing pipeline
Students should explain the pipeline, false positives and false negatives, explicit NO_TARGET and UNSURE behavior, privacy boundaries, and results from more than one environment.