Object dection and speed estimation.
yolov8n
, which means there is wide room for improvement if I use a more powerful one, such as yolov8l
or yolov8x
.dataset details can be found here
patience
value is really low to draw any meaningful conclusion.latest training result (yolov8n).
the model trained mostly on “far view” dataset, such that I tried to use it on this video
Based on these, we can see that: speedX = distanceX * meter_per_pixel * fps
(each frame)
To get the distance, simply store the id
of each car in a dictionary, then compare and update it every frame (or any duration of time).
My initial approach is to use Motion detection + Custom tracking based on distance, but later I decided to use yolov8
for expansion potential and learning purposes.
initial idea of tracking + assigning id in ROI (region of interest)
results of the initial ‘motion detection’ approach