Tracking with yolov5

This implementation is for who need to tracking multi-object only with detector. You can easily track mult-object with your well trained yolov5 model. I used SORT algorithm implementation to track each bounding boxes.

And I added my nobel(maybe) smoothing method. This method reduces the shaking of bounding boxes. You can easily deactivate smoothing method by specifying --no_smoothing option.

I hope this repository can help someone :)

Preparation

This implementation use yolov5 detection results. If you have another trained detector just follow this format.

# file name
[video_name]_[frame_idx].txt

# center_x, center_y, width, height should be normalized with Video Width Height
class_id center_x center_y width height confidence 
...

or just run (if you have trained yolov5 model)

python detect.py  --weights [your model weight]
                  --source  [video path]
                  --save-txt --save-conf --nosave --iou-thres 1 --img 640  # keep this line same

Run

I uploaded a sample dataset in sample_data directory. If you want to use this data, just run

python main.py --show

If you have your own dataset, run

python main.py  --video_path [video path]
                --det_dir    [your yolov5 det files directory]
                --show       [if you want to pre-visualize the results]

Result

You can find your result video and text file under results directory.