Driver assistance systems combine radar and cameras because each sensor is strong exactly where the other is weak. The difficulty is deciding what to believe when they conflict.
Radar measures motion, not identity
Radar transmits radio waves and measures the reflection, giving distance directly and closing speed through the frequency shift of the returned signal.
Those measurements are precise and largely unaffected by darkness, fog, rain or glare, which is why radar underpins adaptive cruise control.
What radar cannot do is say what it has detected. A road sign, a manhole cover and a stationary vehicle can all produce a strong return.
Cameras identify objects but estimate distance poorly
A camera sees shape, texture and colour, allowing software to classify a pedestrian, a cyclist, a lane marking or a traffic sign with reasonable reliability.
Distance from a single camera must be inferred from apparent size and position, which is far less precise than a direct measurement.
Performance degrades in the conditions radar tolerates easily, including low sun straight into the lens, heavy spray and a dirty windscreen.
Stationary objects are the classic disagreement
Radar systems have historically been configured to ignore stationary returns, because reacting to every roadside reflection would produce constant false braking.
That filtering is also why a stationary vehicle in a lane can be treated as background clutter rather than as an obstacle requiring action.
The camera is expected to catch this case by recognising the shape of a vehicle, which places a great deal of weight on classification accuracy.
Fusion has to resolve the conflict
Sensor fusion combines the inputs into one model of the surroundings, weighting each sensor according to its reliability in the current conditions.
Where they agree, confidence is high and the system can act firmly. Where they disagree, the system must choose between a possible false alarm and a possible missed event.
Manufacturers tune that balance conservatively, because unnecessary emergency braking creates its own hazard and rapidly destroys driver trust.
Additional sensors attack the same weakness
Laser scanning produces a direct three-dimensional measurement of shape and distance, which addresses the classification and ranging problems together.
Its cost and packaging have limited adoption, and its performance in heavy precipitation is a genuine engineering constraint rather than a marketing argument.
Higher-resolution radar that can resolve object height is emerging as a cheaper route to the same end, aimed squarely at the stationary object problem.