Traffic-Vehicle-Tracking

Counting no of traffic vehicles using YOLOv5s with DeepSORT algorithm

Script to count the number of instances per vehicle class for a given traffic video clip.

Main components:

• Pytorch
• Opencv
• FFMPEG

Usage(for object tracking):

In order to show the inferences on a video file, for example: traffic_short.mp4, use the following command:

python main.py --input_path traffic_short.mp4 --display

In order to save the tracking inferences in form of video, use the following command:

python main.py --input_path traffic_short.mp4 --save_path out.mp4 --device 0 --display

Usage(only for detection):

In order to show the inferences on a video file, for example: traffic_short.mp4, use the following command:

python main.py --video traffic_short.mp4 --show_detections

To set the custom values of NMS threshold, confidence threshold, use the following command:

python main.py --video traffic_short.mp4 --show_detections --iou_thres 0.15 --conf_thres 0.20 

There is a cutoff distance is being used in the script. It is a threshold distance in pixels beyond which no detection instances will be counted. To use the threshold distance, use the following command:

python main.py --video traffic_short.mp4 --show_detections --distance_thres 300

To save the detection outputs in form of a video instead of displaying them in opencv window, use the - - save_path flag in the following way:

python main.py --video traffic_short.mp4 --save_path detection_output.mp4

External dependencies:

• YOLOV5 [https://pytorch.org/hub/ultralytics_yolov5/]
• Traffic video clip [https://www.youtube.com/watch?v=wqctLW0Hb_0&t=1s]

Additional

FFMPEG command to trim the long 30 minute video downloaded from youtube.

ffmpeg -ss 00:0:01 -i traffic.mp4 -to 00:02:00 -c copy output.mp4