Project that estimates the distance a car is on a road based on the relationship between the real size of the car and the size it appears in the video obtained. It also estimates the lane you are traveling in at any given time based on the angle between the position of the car and the video camera, and you can even guess lane change intentions based on this angle.
It is an attempt to apply the project https://medium.com/@batuhansenerr/object-distance-direction-detection-for-blind-and-low-vision-people-c3f3fd83cbef to the case of detection of distances of cars and traffic lanes roads, despite the differences between both cases.
A standard car width is assumed equal to the width of a parking space (2.3 meters). To avoid the dispersion of the angles with which the lane is estimated, it is only calculated at a distance of 15 meters from the camera
Requirements: all the files that accompany the project and packages that, if missing, can be installed with a simple pip
import numpy as np
import cv2
import time
import math
from ultralytics import YOLO
It is tested by running:
VIDEODetectCarDistanceAndRoadLane.py
best.pt is the model to detect the cars in the video and was created following the instructions that appear in the project https://github.com/ablanco1950/LicensePlate_Yolov8_MaxFilters training yolo with a custom roboflow dataset
References:
The video used: Traffic IP Camera video.mp4 test video, was downloaded from https://github.com/anmspro/Traffic-Signal-Violation-Detection-System/tree/master/Resources
https://es.planetcalc.com/1897/
https://github.com/ablanco1950/LicensePlate_Yolov8_MaxFilters