object_detection.md

Object detection

The inference results are provided in the eCAL topic object_detection as JSON data. The following example show how an object could look like:

{
    "iso_time": "2024-11-05T14:23:07.263719",
    // Class labels for each bounding box
    "class_ids": [
        0.0 // person
    ],
    // Confidence scores for each box
    "confidences": [
        0.944169819355011 // confidence person
    ],
    // Boxes in [x1, y1, x2, y2] format
    "xyxy": [
        [
            141.6815185546875,  // x1
            182.14732360839844, // y1
            564.3827514648438,  // x2
            479.19677734375     // y2
        ]
    ]
}

Note: The original image size is 480x640 (width x height).

The following table describes the class ids with the corresponding lables:

class_id	label
0	person
1	bicycle
2	car
3	motorcycle
5	bus
6	train
7	truck
9	traffic light
11	stop sign
12	parking meter

Note: There is only pre-defined confidence filter applied. Find the best confidence threshold within your specific application and use-case.