Documentation

We have implemented two different approaches to car detection and counting in this repository.

---

Approach A: Reference Points

To use this approach with the default configuration simply run the following command

python  car_detection_refPoints.py

This will start the camera and take the first image for setting of reference Points. On this image click on two points on each lane. The first point ("point 1") of each lane should be reached before the second one ("point 2") by a car that is driving on the lane. The reference points should not be on the edge of the image but at a place were the entire car is completly visible. The reference point should only be reachable via one lane.

configuration

Name	short	type	range	default	description	required
--confidence	-c	float	0 - 1	0.47	confidence level at which a detection is being counted	no
--fromfile	-f	str	-	''	File from which the videoframes are being read. If '' The camera is used as input	no
--lanepoints	-p	int	1 - 2	2	number of points to reference a lane	no
--lanes	-l	int	1 - 10	2	number of lanes in the video	no
--overlay	-o	-	-	-	deactivate the overlay	no

---

Approach B: Similarity

To use this approach simply run the following command

python  car_detection_similarity.py

This approach will count cars overall and not differentiate between lanes