You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In radars with object output, there are cases that multiple detection results are obtained from one object, especially for large vehicles such as trucks and trailers.
15
15
Its multiple detection results cause separation of objects in tracking module.
16
16
Therefore, by this package the multiple detection results are clustered into one object in advance.
17
17
18
-
### Detail Algorithm
18
+
### Algorithm
19
19
20
-
- Sort by distance from `base_link`
20
+
-1.Sort by distance from `base_link`
21
21
22
22
At first, to prevent changing the result from depending on the order of objects in DetectedObjects, input objects are sorted by distance from `base_link`.
23
23
In addition, to apply matching in closeness order considering occlusion, objects are sorted in order of short distance in advance.
24
24
25
-
- Clustering
25
+
-2.Clustering
26
26
27
27
If two radar objects are near, and yaw angle direction and velocity between two radar objects is similar (the degree of these is defined by parameters), then these are clustered.
28
28
Note that radar characteristic affect parameters for this matching.
@@ -32,13 +32,13 @@ For example, if resolution of range distance or angle is low and accuracy of vel
32
32
33
33
After grouping for all radar objects, if multiple radar objects are grouping, the kinematics of the new clustered object is calculated from average of that and label and shape of the new clustered object is calculated from top confidence in radar objects.
34
34
35
-
- Fixed label correction
35
+
-3.Fixed label correction
36
36
37
37
When the label information from radar outputs lack accuracy, `is_fixed_label` parameter is recommended to set `true`.
38
38
If the parameter is true, the label of a clustered object is overwritten by the label set by `fixed_label` parameter.
39
39
If this package use for faraway dynamic object detection with radar, the parameter is recommended to set to `VEHICLE`.
40
40
41
-
- Fixed size correction
41
+
-4.Fixed size correction
42
42
43
43
When the size information from radar outputs lack accuracy, `is_fixed_size` parameter is recommended to set `true`.
44
44
If the parameter is true, the size of a clustered object is overwritten by the label set by `size_x`, `size_y`, and `size_z` parameters.
@@ -51,28 +51,75 @@ Note that to use for [multi_objects_tracker](https://github.com/autowarefoundati
51
51
For now, size estimation for clustered object is not implemented.
52
52
So `is_fixed_size` parameter is recommended to set `true`, and size parameters is recommended to set to value near to average size of vehicles.
|`angle_threshold`| double | Angle threshold to judge whether radar detections come from one object. [rad]| 0.174 |
71
-
|`distance_threshold`| double | Distance threshold to judge whether radar detections come from one object. [m]| 4.0 |
72
-
|`velocity_threshold`| double | Velocity threshold to judge whether radar detections come from one object. [m/s]| 2.0 |
73
-
|`is_fixed_label`| bool | If this parameter is true, the label of a clustered object is overwritten by the label set by `fixed_label` parameter. | false |
74
-
|`fixed_label`| string | If `is_fixed_label` is true, the label of a clustered object is overwritten by this parameter. | "UNKNOWN" |
75
-
|`is_fixed_size`| bool | If this parameter is true, the size of a clustered object is overwritten by the label set by `size_x`, `size_y`, and `size_z` parameters. | false |
76
-
|`size_x`| double | If `is_fixed_size` is true, the x-axis size of a clustered object is overwritten by this parameter. [m]| 4.0 |
77
-
|`size_y`| double | If `is_fixed_size` is true, the y-axis size of a clustered object is overwritten by this parameter. [m]| 1.5 |
78
-
|`size_z`| double | If `is_fixed_size` is true, the z-axis size of a clustered object is overwritten by this parameter. [m]| 1.5 |
These parameter are thresholds for angle, distance, and velocity to judge whether radar detections come from one object in "clustering" processing, which is written in detail at algorithm section.
76
+
If all of the difference in angle/distance/velocity from two objects is less than the thresholds, then the two objects are merged to one clustered object.
77
+
If these parameter is larger, more objects are merged to one clustered object.
78
+
79
+
These are used in `isSameObject` function as below.
0 commit comments