Notes.txt

1] Create a Yolo-Weights Folder where U will execute a code and export pytorch(.pt) files into
    openvino format so as to get openvino model folders for specific yolo package

2] Created Yolo-webCam folder to connect our Yolo models to our webcam
   But the problem is that for all this cvzone or cv2 is using CPU for processing
   (If we directly use .pt files)
   so now we gotta run this with GPU thats we previously exported in openvino format
   (As openvino supports intel GPUs)

3] Now we create car counter project in a separate folder
   Till now our whole video was being scanned by YOLO but we dont want that
   We just want it to scan a specific part of the road to detect cars
   So we paste our video on canva to create a mask for the video

4] Now we gotta import this mask.png in our project file
   and we overlay this mask on our img (vid) and pass this image as a result to our Yolo model
   So now our YOLO detects cars only ina a specific part

5] Now the YOLO is only detecting the cars but not tracking them
    Which means that it detects each car in each frame separately
    So we use a tracker named SORT designed by abewley

6] To use this we create a line on our img with cv2 then also create a dot on each identified OBj
   So when these dots pass through this line they are counted

********** SORT THEORY ***********

max_age
This parameter determines how long a track can remain alive without receiving new detections.
Tracks that do not receive detections for more than max_age frames are considered inactive and are removed from the tracking process.
Increasing max_age allows tracks to persist longer without updates, which can be helpful in scenarios where objects may momentarily disappear from the scene and then reappear.

min_hits:
This parameter specifies the minimum number of hits required to establish a track for an object.
When an object is first detected, it needs to receive detections in consecutive frames at least min_hits times to be considered a valid track.
Setting a higher value for min_hits can help filter out false positives and ensure that only reliably detected objects are tracked.

IoU Threshold (Intersection over Union Threshold):
Imagine two boxes: one representing a detection from YOLO, and the other representing a track maintained by the SORT tracker.
IoU measures how much these two boxes overlap. It's the ratio of the area of overlap between the boxes to the area of union.
The IoU threshold determines how much overlap is required for the tracker to associate a detection with an existing track.
If the IoU between a detection and a track exceeds the threshold, the tracker considers them as the same object.
A lower IoU threshold means even slight overlap is enough for association, potentially leading to more matches but also more false positives.
A higher IoU threshold requires more significant overlap, reducing false associations but potentially missing some valid matches.
Raising the iou_threshold requires a stricter overlap between detections and tracks, reducing the likelihood of false associations but potentially leading to more missed detections.

In summary, max_age controls the lifespan of tracks, min_hits determines the reliability of detections required to establish a track, and iou_threshold governs the association between detections and existing tracks based on their spatial overlap. Adjusting these parameters allows for fine-tuning the performance of the tracker based on specific tracking requirements and characteristics of the input data.