This is a basic framework for a functional object detection combined with multiple object tracking.
The object detection is performed using tensorflow object detection api while the tracking part is performed using opencv tracking.
| Argument | Description | Example | Default |
|---|---|---|---|
-h,--help |
show help message | python3 object_detection_tracking.py -h |
|
--version |
check version | python3 object_detection_tracking.py -v |
|
--visualize |
Whether to visualize the results in every frame. | python3 object_detection_tracking.py --visualize |
False |
--score-threshold |
The score above which bboxes are taken into consideration. | python3 object_detection_tracking.py --score-threshold=0.7 |
0.5 |
-v,--video-input-path |
The path to the video to be processed | python3 object_detection_tracking.py -v="path-to-video" |
Required |
--video-output-path |
Where to write the output video | python3 object_detection_tracking.py --video-output-path="path-to-output-video" |
(input_video_name)_detected(.ext) |
-m,--model-path |
Path to the frozen model to be used | python3 object_detection_tracking.py --model-path="path-to-model/frozen_inference_graph.pb" |
model/frozen_inference_graph.pb |
-l,--labels-map-path |
Path to the labels map to be used. | python3 object_detection_tracking.py --labels-map-path="path-to-labels-map/labels.pbtxt" |
model/label_map.pbtxt |
--detection-rate |
The rate of detection, it perform 1 detection every detection-rate value. | python3 object_detection_tracking.py --detection-rate=10 |
5 |
The simpler way is to provide a path to a video to be processed using the provided model.
python3 object_detection_tracking.py --video-input-path="path-to-input-video/test_video.mp4"
-
this way a new output video will be created named:
test_video_detected.mp4. This can be changed using the arguments above. -
In this video all detected objects with confidence score >= 0.5 will be added a bounding box around them
-
Object tracking will be perfomed every 5 frames for each of the above detected objects.
-
After the tracking a new object detection will be performed