Assigning unique ID to each bounding box to track it's movement

[Passionatecricketer]

Hi,

I have a custom object detector using faster rcnn but now I want to track every object's movement in a video and get a count on the basis of movement from left to right. How can I assign an unique id to each bounding box and also bypass the occlusion issue. I know there is deep sort algorithm but I want to know if anything is possible through this API only or anything that have an application with faster rcnn

Thanks in advance!

[luojueling]

i have same problem

[mohammadakz]

@Passionatecricketer @luojueling

Hi, I believe you can NOT use Object detection API for tracking directly, and you have to use some other trackers such as deep sort. A potential solution could be using the repository below to count the detected objects if the overall goal is to have a number of objects existing on the scene: https://github.com/ahmetozlu/tensorflow_object_counting_api if not, I have implemented centroid tracking with object detection, and the results are acceptable. You can find that online 156 in the repository below: https://github.com/mohammadakz/Nested_PPE_detection/blob/master/nested_detection.py

I hope that helps