Darknet(YOLO) + tracker

Darknet is an open source Object detecter written in C and CUDA. It’s fast, easy to install, and supports CPU and GPU computation.

but it can’t tracking the thing, which detected by yolo.
So we have created a system that can track each object by processing the information of the objects obtained from YOLO.

This is what we different between YOLO original and our project.

In the existing YOLO, only the location and class of objects can be known, but our system can distinguish each object.
To make this possible, we used various methods such as ORB feature matching, histogram matching, and object velocity prediction.

Our system consists of two algorithms.

1. one view object tracker.
2. muti view object tracker.

in one view object tracker, it’s distinguishing and tracking objects in only one view. this tracker initiate with YOLO frame, which have object’s infomation.

First, the tracker is initialized based on the object information obtained from the YOLO frame. We used the MOSSE tracker provided by OPENCV. If the MOSSE tracker initialization completes successfully, the remaining four frames track the object with the MOSSE tracker. If the MOSSE tracker fails to track due to some reason, the following three methods are used to compensate.

1. 

<!—