A stopped vehicle, a guardrail and an overhead road sign can all produce strong radar returns. However, only one of them may require emergency braking.
This illustrates one of the most important challenges in automotive radar perception: detecting a stationary reflection is not the same as determining whether it is a relevant road hazard.
Automotive radar must separate objects that occupy the driving path from the much larger number of stationary structures surrounding the road. The sensor may detect many valid returns, but the ADAS controller still needs to determine which targets should influence Forward Collision Warning (FCW), Automatic Emergency Braking (AEB) or automated-driving decisions.
This article explains why stationary targets are difficult, how common road situations affect radar output, and how long-range and 4D imaging radars can support more reliable perception.
Stationary Does Not Mean Invisible to Radar
It is sometimes claimed that automotive radar cannot detect stationary objects. That statement is too broad.
FMCW automotive radar can measure the range and angle of stationary reflectors. Depending on the radar architecture, installation, processing and operating situation, it may also generate point-cloud or tracked-target information for stationary objects.
The real difficulty is relevance.
A radar mounted on a moving vehicle can receive reflections from:
- Stopped vehicles;
- Guardrails;
- Road signs;
- Bridges;
- Tunnel walls;
- Lamp posts;
- Fences;
- Parked vehicles;
- Roadside buildings;
- Construction barriers;
- Objects lying on the road.
Many of these objects are genuinely stationary and correctly detected. However, most do not require a driver warning or braking response.
The perception system therefore needs to answer a more difficult question:
Is the stationary target physically located in the vehicle’s expected driving path, and does it represent a real collision risk?
Why Doppler Velocity Alone Is Not Enough
Radar measures the radial component of relative velocity. This measurement is extremely useful for distinguishing approaching and receding vehicles, but stationary-road scenes require additional interpretation.
When the host vehicle is stopped, road infrastructure may have little or no relative Doppler velocity. Returns from a stopped car, guardrail and road sign can therefore occupy similar low-velocity regions.
When the host vehicle is moving, stationary objects appear to move relative to the radar because the radar itself is moving. Their measured radial velocities vary according to:
- Host-vehicle speed;
- Target position;
- Radar installation angle;
- Road curvature;
- Vehicle yaw rate;
- Target angle relative to the radar beam.
Consequently, a simple rule such as “zero velocity means stationary” is not sufficient for every driving condition. The system may need host-vehicle motion information and coordinate compensation to estimate whether an object is stationary relative to the road.
Velocity remains valuable, but it must be interpreted together with target range, angle, height, track history and driving-path information.
The Road Contains More Stationary Clutter Than Hazards
A moving vehicle is often easier to distinguish from a static background because its motion differs from the surrounding infrastructure. A stationary hazard must be identified within that background.
Consider a radar travelling along a highway. It may continuously receive returns from a guardrail extending for hundreds of metres. The guardrail is real, close to the vehicle and strongly reflective, but it normally follows the road boundary rather than crossing the driving path.
A stopped vehicle in the same lane may initially be surrounded by these returns. The perception system must determine that the stopped vehicle has a different position, shape and track pattern and that it occupies the host vehicle’s projected path.
The problem becomes more difficult when:
- The road curves;
- Lane markings are unclear;
- A large vehicle blocks part of the view;
- The stopped vehicle is close to a guardrail;
- The radar is installed behind a complex bumper;
- The road is inside a tunnel;
- Multiple stationary objects are positioned close together.
This is why maximum detection range alone cannot describe stationary-target performance. A useful system must also provide stable spatial information that allows relevant targets to be separated from background structures.
Common Case 1: A Stopped Vehicle in the Driving Lane
A vehicle has stopped because of congestion, a breakdown or an accident. The host vehicle approaches at highway speed.
From a safety perspective, this is one of the most important stationary-target cases. The perception system must identify the stopped vehicle early enough to support warning or braking decisions.
The difficulty is not simply receiving a reflection. The system must determine that:
- The target is in or near the host vehicle’s path;
- It is not a roadside vehicle outside the lane;
- The track is stable over consecutive radar updates;
- Its position agrees with the road and lane geometry;
- The closing risk results mainly from host-vehicle motion;
- The target is not a temporary ghost created by multipath.
A camera can provide lane and object-category information, while radar provides range, angle and relative-motion measurements. Combining these inputs can help the ADAS controller determine whether the object is a relevant stopped vehicle.
Radar hardware supplies perception data; the final FCW or AEB decision normally depends on the complete vehicle sensing and control architecture.
Common Case 2: A Guardrail Beside a Curved Road
Guardrails can produce strong and continuous radar returns. On a straight road, they may be relatively easy to associate with the road boundary. On a curve, however, sections of the guardrail may appear geometrically closer to the vehicle’s forward direction.
If target selection relies only on range, a nearby guardrail return could be mistaken for an object in the driving path. If static returns are filtered too aggressively, the system could also suppress a genuine stopped vehicle located close to the guardrail.
A practical perception system may consider:
- The continuity of returns along the road edge;
- The lateral distribution of radar points;
- Road curvature;
- Host-vehicle steering and yaw rate;
- Whether the return forms an extended roadside structure;
- Whether a separate object track remains in the lane.
This case demonstrates why road geometry and multi-frame spatial patterns are important.
Common Case 3: An Overhead Sign or Bridge
An overhead sign, bridge or gantry may be directly ahead of the vehicle and generate a strong return. In a conventional two-dimensional representation, the target could appear to occupy the forward path even though the vehicle can safely pass underneath it.
Elevation information can improve this situation.
A 4D imaging radar measures not only range, azimuth and velocity but also elevation. Higher-density point-cloud information can help the perception system estimate whether returns belong to:
- A vehicle on the road;
- A sign above the road;
- A bridge structure;
- A traffic-light assembly;
- A low obstacle that may block the vehicle.
Elevation measurement does not automatically solve every classification problem. The controller still needs to consider radar installation, vehicle height, point-cloud quality and the geometry of the detected structure.
Nevertheless, the vertical dimension provides information that a traditional range-and-azimuth target list may not describe sufficiently.
Common Case 4: Debris on the Road
A tyre, box or other piece of debris may be stationary and positioned directly in the lane. Compared with a passenger vehicle, it may have:
- A smaller physical size;
- A lower height;
- A weaker or less stable radar cross-section;
- Fewer point-cloud returns;
- Greater sensitivity to viewing angle.
This makes road debris a different problem from detecting a stopped vehicle.
The system must avoid treating every small reflection as an emergency while still retaining potentially relevant low-profile obstacles. In many implementations, this requires cooperation between radar, camera and other perception sensors.
When discussing debris detection, it is important not to assume that a radar will detect every object at its maximum specified vehicle range. Detection performance depends on target size, material, orientation, radar cross-section, installation and environmental conditions.
Common Case 5: Stationary Targets Inside a Tunnel
Tunnels are particularly challenging because they contain many reflective surfaces:
- Side walls;
- Ceiling structures;
- Ventilation equipment;
- Signs;
- Lighting assemblies;
- Emergency cabinets;
- Repeating structural features.
Multiple reflections can produce clutter or ghost targets that do not correspond directly to one physical object. A stopped vehicle may also appear against a dense background of wall and ceiling returns.
Useful radar data in this environment requires stable range, angle and tracking performance. Point-cloud density, spatial resolution and multipath filtering can help the perception software distinguish road users from the tunnel structure.
Testing should include more than an empty tunnel. Relevant scenarios include:
- A stopped passenger car near the wall;
- A motorcycle stopped behind a larger vehicle;
- Slow-moving congestion;
- Vehicles changing lanes;
- Entry and exit lighting transitions;
- Curved tunnel sections;
- Repeating overhead structures.
How Automotive Systems Evaluate Stationary Targets
The exact processing architecture varies, but several types of information are commonly useful.
Range, Angle and Elevation
Range determines how far away the reflection is. Azimuth helps determine whether it is inside or outside the projected path. Elevation can help distinguish road-level objects from overhead structures.
Host-Vehicle Motion
Vehicle speed, yaw rate and steering information can help compensate for the radar’s own movement and relate radar targets to the road coordinate system.
Multi-Frame Tracking
A persistent track is generally more useful than a single radar point. Tracking allows the system to examine whether the target position remains physically consistent over time.
Driving-Path Association
An object can be close without being dangerous. The controller needs to determine whether the target overlaps the current or predicted driving path.
This assessment becomes especially important on curves and during lane changes.
Point Clustering and Spatial Distribution
A group of radar points may represent one vehicle, part of a guardrail or several adjacent objects. Clustering and spatial analysis help transform individual reflections into more useful scene information.
Sensor Fusion
Radar provides strong range and relative-velocity information. Cameras can add lane boundaries, object categories and visual context. Other sensors may provide additional geometry.
Fusion does not mean that one sensor is always correct. It means the system can compare complementary evidence before making a safety-related decision.
These processing functions may be implemented inside a radar, an ADAS ECU or a domain controller. Integrators should confirm the division of responsibility from the radar datasheet, communication protocol and system architecture.
CTLRR-220PRO for Long-Range Forward Perception
The CTLRR-220PRO is a 77GHz long-range millimeter-wave radar intended for forward ADAS applications.
According to its technical datasheet, it provides:
- Detection range up to 260 metres;
- 4T4R FMCW architecture;
- Distance, velocity and angle measurement;
- Point-cloud and track data modes;
- Support for 40 tracked-target outputs;
- Up to 1,024 point-cloud points;
- 50 ms refresh period;
- CAN and CAN-FD interfaces;
- Support for integration with vision-based ADAS systems;
- Application support for FCW, AEB and ACC.
These characteristics provide an information source for long-range target detection and tracking. For stationary-target use cases, however, maximum range should not be interpreted as a guaranteed stationary-object detection distance for every target.
A stopped passenger car, small road object and roadside sign have different radar characteristics. Final performance must be assessed using representative targets, the production bumper, the intended installation position and the complete ADAS software.
The CTLRR-220PRO can supply radar perception data, while the vehicle controller may remain responsible for lane association, path prediction, object relevance and final warning or braking decisions.
CTLRR-540 for Higher-Resolution Spatial Understanding
The CTLRR-540 is a 77GHz 4D imaging front radar designed for L2+ and more advanced perception applications.
Its product manual specifies:
- 6T8R radar architecture;
- FMCW and DDMA operation;
- Vehicle detection up to 360 metres;
- Up to 2,048 point-cloud points per frame;
- Tracking of up to 256 targets;
- Point-cloud and track output;
- Azimuth and elevation measurement;
- 50 ms data period;
- CAN and 100 Mbps Ethernet interfaces;
- Application support for AVP, NOA, SLAM, AEB, ACC and FCW.
The elevation dimension and denser point cloud can provide more spatial information for differentiating road-level targets from structures above or beside the road.
The product manual also identifies high-altitude passable targets, tunnels, overpasses, billboards, intersection structures and ordinary road sections as relevant imaging scenarios.
This does not mean that every CTLRR-540 installation automatically produces a final classification such as “safe overhead sign” or “hazardous stopped vehicle.” Depending on the selected output and system design, additional clustering, target classification, free-space estimation, path association or sensor fusion may be performed by an external controller.
Specific point-cloud fields, target attributes and interface behavior must be confirmed from the applicable communication protocol.
How to Validate Stationary-Target Performance
Stationary-target validation should use controlled and repeatable scenarios. A useful test plan can include the following steps.
- Verify the Installation
Check:
- Radar mounting height;
- Yaw, pitch and roll angles;
- Bracket stiffness;
- Radar position relative to the vehicle centreline;
- Bumper clearance;
- Metal or conductive parts inside the field of view;
- Connector and communication status.
Incorrect installation can make a software problem appear to be a radar-performance problem.
- Establish a Ground-Truth Reference
Record the actual position and dimensions of each target. Synchronised video, surveyed markers or another validated reference sensor can help compare the radar output with the physical scene.
- Test Different Stationary Targets
Testing should include more than one stopped passenger car:
- Passenger car;
- Motorcycle;
- Truck;
- Metal barrier;
- Plastic road barrier;
- Road debris of different sizes;
- Roadside parked vehicle;
- Overhead structure.
- Change the Road Geometry
Repeat tests on:
- Straight roads;
- Curves;
- Slopes;
- Multi-lane highways;
- Tunnels;
- Roads with guardrails;
- Roads with overhead signs.
- Vary Host-Vehicle Conditions
Use different:
- Host speeds;
- Approach angles;
- Lane positions;
- Steering inputs;
- Traffic densities;
- Target occlusion conditions.
- Evaluate More Than Detection
Useful evaluation results include:
- First valid detection distance;
- Track-confirmation distance;
- Position stability;
- Track ID continuity;
- False-target behavior;
- Path-association accuracy;
- Target loss and reacquisition;
- Warning activation position;
- Behavior near guardrails and overhead structures.
A radar return appearing once at long range is not equivalent to a stable target that an ADAS function can safely use.
Common Integration Mistakes
Removing All Low-Velocity Targets
Aggressive filtering may reduce clutter, but it can also remove stopped vehicles and other relevant hazards.
Treating Every Stationary Return as a Hazard
This can cause warnings or braking responses for guardrails, signs and roadside structures.
Ignoring Ego-Motion Compensation
Host speed and turning motion affect the apparent movement and position of stationary infrastructure.
Using Maximum Range as a Guaranteed Hazard Range
Specified vehicle-detection range does not automatically apply to small, low or weakly reflective objects.
Conclusion
Stationary targets are difficult for automotive radar because the road contains many more stationary reflections than genuine stationary hazards.
The radar may correctly detect a stopped vehicle, guardrail, bridge and road sign at the same time. The ADAS system must then determine which object occupies the driving path, which belongs to the surrounding infrastructure and which requires a warning or braking response.
Long-range radar such as the CTLRR-220PRO can provide forward target measurement and tracking information for FCW, AEB and ACC integration. The CTLRR-540 adds denser 4D point-cloud information and elevation measurement for more detailed spatial understanding.
The final result depends on more than the radar specification. Correct installation, host-motion compensation, target tracking, path association, sensor fusion and representative vehicle testing are all necessary.
For product selection, communication documentation or automotive radar integration support, contact ZLYRADAR with your target scenarios, vehicle type, installation position, required interface and ADAS functions.
