No uniform detection ranges of a camera within entire FoV.

I am using driving scenario generator for camera based detections. It is observed that when an object is placed on the center line of FoV, the detection range is limited to a short range (in meters). But, when object is at the edge of the FoV lines, then camera is detecting object till very long range.

The sensor parameters used

The camera detections at 60m and 80m for pedestrian

Please do let me know why uniform detection within entire FOV of camera is not appearing.

Answer 1

[Akanksha]

Hey Saikumar,

The reason the camera is not detecting objects evenly across its entire field of view (FoV) has to do with how the camera sees the world. Even though the detection range is technically the same across the FoV, the way objects appear in the image can vary a lot depending on where they are.

Here’s what’s likely happening:

1. Objects in the centre look smaller When something is directly in front of the camera and far away, it appears really small in the image. If it is too small ie smaller than the minimum size you’ve set for detection (15×15 pixels)—the camera just would not pick it up.
2. The camera is angled downward Your camera has a 10° downward pitch, which means it’s looking more at the ground in front of the vehicle. So, objects far away in the centre might be too low in the image or too small to detect.
3. Objects at the edges can appear larger Surprisingly, objects that are off to the side (at the edge of the FoV) can actually appear bigger in the image at the same distance. That’s because of how the camera projects the 3D world onto a 2D image—it stretches things more at the edges.

So even though the camera’s range is the same across the whole view, the way objects are projected into the image makes a big difference in whether they’re detected.

Below are few things you could try to improve camera detection :

1. 1. Lower the object size threshold: Change it from 15×15 to 10×10 pixels so the camera can detect smaller or distant objects.
2. Reduce the camera’s downward tilt: Try changing the pitch from 10° to 5° to help it see farther ahead.
3. Widen the field of view: Adjust the focal length to make object sizes more consistent across the image.

Enclosed are the documentation along with the subsection that will back the solution provided above and help you for further reference :

https://www.mathworks.com/help/releases/R2023a/driving/ug/model-vision-sensor-detections.html?searchHighlight=Model+Vision+Sensor+Detections&s_tid=doc_srchtitle – Refer “Pedestrian and Vehicle Detection”

https://www.mathworks.com/help/releases/R2023a/driving/ref/visiondetectiongenerator.html?searchHighlight=Vision+Detection+Generator&s_tid=doc_srchtitle - Refer subsection “Minimum detectable image size of an object”  and “Pitch angle of sensor mounted on ego vehicle (deg)”

Hope this resolves your doubt.