Yolo wasn't detecting well some classes (mainly pedestrians) on the stitched/panorama image. To improve the outcome, I used the same pedestrian model and positioned it precisely at the image boundary zone, as it is the most challenging scenario during the stitching process.
Left camera image
Right camera image
The following images show the resulted panorama. Every class detection was tested in the boundary zone.
Full panorama
Front car view
Lateral car view
Pedestrian
Rider
Truck front view
Truck lateral view
Pedestrian
Car