Analytics & ML · Study deck

Sensor Fusion Foundations

Use a short fusion review.

Data Dora is your guide for this deck.

datafusion

Data Dora, the module guide, in a scene from this chapter.

After studying this chapter

Learning objectives

You will be able to:

Explain: The contract starts with a state sentence that is narrow enough to test. "Estimate the temperature field in a greenhouse bay every 60 seconds" is reviewable; "understand the greenhouse" is not.
Explain: The durable starting point is not an algorithm name; it is a fusion contract that records the state, sensor roles, fusion level, timing, uncertainty, conflict policy, validation target, and degraded modes.
Explain: A calibration table, floor plan, map, equipment database, or known user context can make a fused estimate more accurate or more specific than one sensor could make alone.

Major section

Start With the Story

Recheck when a source, mount, clock, or use changes.

One says the room is warm.
The other says it is cool.
A controller should not average them at once.
One sensor may be near a heater.

Major section

Start With the Story (continued)

This first view uses a small set of sources with known roles.

Their times may not match.
Data fusion means combining evidence from more than one source.
The aim is a more useful result, not merely a new number.
Two values from different moments may describe different events.

One timestamp through a redundant-fusion rule: gate on the median, average the survivors, publish the exclusion with the value.

Major section

Start With the Story (continued)

The mathematical gist: An 8.0° accelerometer tilt means $a_x=9.81\sin(8.0°)=1.37$ m/s²; a catalog-typical MEMS mass and spring turn that into 5.12 nm of motion.

Two labels may use different meanings.
A missing value should stay visible.
You may select the best source.

Major section

Start With the Story (continued)

You may use a weighted blend.

You may let one source correct another.
Under the Hood examines uncertainty, weighting, timing, and why more inputs do not always mean a better answer.
A greenhouse controller should align two temperature readings in time and context before combining them.

Major section

Fusion Combines Evidence

Sensor fusion combines measurements, features, or decisions from more than one source so an IoT system can estimate state with a clearer evidence boundary than any single sensor can provide.

A fusion design should name what each sensor contributes, how time is aligned, how uncertainty is represented, and what the system does when sensors disagree or disappear.

The same segmentation step producing either a filtered image or a labelled list of connected components

Major section

Fusion Combines Evidence (continued)

The most common beginner mistake is treating fusion as simple averaging.

Averaging can reduce random noise when sensors measure the same quantity with independent errors, but many IoT systems need complementary evidence.
A GPS receiver provides slow absolute position; an IMU provides fast motion changes but drifts.
Fusion level is the next choice.

Cooperative fusion at work: an attitude team and a height team turn four commodity sensors into a track no datasheet lists.

Major section

Fusion Combines Evidence (continued)

Feature-level fusion waits until useful features have been extracted.

A camera gives rich scene evidence but is sensitive to lighting; radar can keep range evidence under different visual conditions.
Fusion is the discipline of combining those strengths while keeping their failure modes visible.
Decision-level fusion combines detector outputs when raw evidence cannot be shared.

Major section

Fusion Combines Evidence (continued)

A practical input-output view makes the same choice more concrete.

If you only need the intuition, this layer is enough: fusion is a documented evidence contract.
The review question is always what evidence was lost, summarized, or made harder to replay at that level.
Classical multisensor fusion also includes related information, not only live sensor streams.

One monitoring question fused at four different stages -- and what remains replayable after each choice.

Major section

Fusion Combines Evidence (continued)

Redundant fusion combines sources that observe the same target so confidence or fault tolerance can improve.

The contract starts with a state sentence that is narrow enough to test. "Estimate the temperature field in a greenhouse bay every 60 seconds" is reviewable; "understand the greenhouse" is not.
Once the state is named, each sensor gets a role: one may provide absolute correction, another may provide fast relative change, and another may act as a fault witness when the first two disagree.
The benefit of multiple sensors should be stated in that same evidence language.

Major section

Fusion Combines Evidence (continued)

This difference carries the narrative from representation to evidence: the feature branch is easier to classify but cannot replay every pixel-level choice made before the handoff.

A calibration table, floor plan, map, equipment database, or known user context can make a fused estimate more accurate or more specific than one sensor could make alone.
The review record should therefore name every source of evidence and whether it is measured now, looked up from an associated database, or assumed from the deployment model.
The cooperative case rewards a closer look, because it is the only relationship of the three that manufactures a measurement no input owns.

Major section

Fusion Combines Evidence (continued)

None of those benefits is automatic, so the design still has to prove independence, placement, calibration, and degraded-mode behavior.

The three-panel diagram in Figure: How the sources relate to each other decides contrasts three relationships because each supports a different claim about confidence, coverage, or newly inferred information.
The first practical challenge is often not the algorithm but making the inputs comparable.
Source relationship says why evidence is combined; fusion level says what representation is combined.

How the sources relate to each other decides what the fused result can prove -- the same three relationships this section defines in prose.

Major section

Roles and Complementary Filter · Fusion Quality Constraints

For a small IMU example, the state may be device tilt angle.

The accelerometer can estimate gravity direction over longer intervals, but it is noisy during vibration or acceleration.
interpretation: The gyro carries most of the fast motion for this update.
Fusion improves decisions only when the inputs carry useful, timely, and sufficiently independent evidence.

Real redundant sensors mostly agree -- but a shared-bias outlier is exactly the risk the independence assumption below is checking for.

Major section

Summary · Key Takeaway

Sensor fusion improves IoT estimates by combining evidence from sensors that see different parts of the system state.

The durable starting point is not an algorithm name; it is a fusion contract that records the state, sensor roles, fusion level, timing, uncertainty, conflict policy, validation target, and degraded modes.
Complementary filters, Kalman filters, particle filters, weighted estimates, and decision voting are useful only when those boundaries are explicit.
Fusion is trustworthy when it keeps uncertainty visible.

Deck summary

Key takeaways

Recheck when a source, mount, clock, or use changes.

This first view uses a small set of sources with known roles.
The mathematical gist: An 8.0° accelerometer tilt means $a_x=9.81\sin(8.0°)=1.37$ m/s²; a catalog-typical MEMS mass and spring turn that into 5.12 nm of motion.
You may use a weighted blend.
Sensor fusion combines measurements, features, or decisions from more than one source so an IoT system can estimate state with a clearer evidence boundary than any single sensor can provide.

Retrieval practice

Recall check 1 of 4

Data Dora says: answer from memory, then check your reasoning.

Q1Why is sensor fusion more than averaging several sensor readings?

ABecause taking the average reduces the need to calibrate redundant sensors before combining their readings.

BBecause fusion defines sensor roles, timing, uncertainty, conflict handling, and validation boundaries.

CBecause fusion removes the need to test sensor failures, drift, stale data, or calibration.

DBecause production fusion should first learn a neural model that maps the combined readings to a state estimate.

Show answer

Answer: B Fusion is a structured evidence contract, not a generic average.

Retrieval practice

Recall check 2 of 4

Data Dora says: answer from memory, then check your reasoning.

Q2In the worked complementary-filter example, why does the gyro prediction receive most of the weight for one short update?

ABecause the gyro handles short updates while the accelerometer slowly corrects toward gravity.

BBecause the accelerometer is permanently disabled once gyroscope samples are available.

CBecause alpha alone proves the fused angle is accurate without motion validation.

DBecause this complementary filter needs GPS data before it can estimate tilt.

Show answer

Answer: A A complementary filter blends fast relative evidence with slower corrective evidence.

Retrieval practice

Recall check 3 of 4

Data Dora says: answer from memory, then check your reasoning.

Q3Sensors are hard to synchronize, and only compact local summaries may cross the network. Which fusion level deserves review?

ARaw-data fusion with no timing plan.

BDecision-level fusion before any local analysis.

CFeature-level fusion at a shared node.

DState-level fusion without defining the state.

Show answer

Answer: C Features can reduce transferred data while keeping a shared basis for fusion.

Retrieval practice

Recall check 4 of 4

Data Dora says: answer from memory, then check your reasoning.

Q4Why can fusing two sensors with the same hidden bias make a system worse?

ABecause adding a second sensor increases the random noise in the combined estimate.

BBoth inputs can share the same bias while the fused output looks certain.

CBecause fusion must wait until only decision labels are available.

DBecause timestamps are irrelevant once two readings have been combined.

Show answer

Answer: B Fusion quality depends on the error relationship between inputs, not just the number of inputs.

Print reference

Answers

Answer key.

B · Fusion is a structured evidence contract, not a generic average.
A · A complementary filter blends fast relative evidence with slower corrective evidence.
C · Features can reduce transferred data while keeping a shared basis for fusion.
B · Fusion quality depends on the error relationship between inputs, not just the number of inputs.