AI object tracking

Object tracking software

Tracking an object is crucial at numerous places such as airports, assembly lines, or stores. This task might seem to be simple because humans, as well as animals, do it effortlessly, but in fact, tracking objects in a real-world environment require very sophisticated approaches. Almost all recognition tasks that have video as a source include object-tracking technology. After the object is detected, you need a tracking module that will assign an ID to this object and will follow it in all subsequent frames.

AI object tracking water background

The simplest object tracking software relies just on the object’s coordinates. The detector provides a tracker with the object’s centroid in each frame. With the basic assumption that frame-to-frame centroids would move only a little bit, the object-tracking algorithm would assign IDs by looking at relative distances only. This object tracker works fast and relatively well in the refined environment when all objects spaced apart from each other. With more objects in a frame, this approach fails as it starts switching IDs between different objects.

The next level of object tracking technology includes speed and movement direction. By adding these two parameters to the object’s coordinates this tracker is capable of predicting the position of this object in the upcoming frame. If actual coordinates are close to the predicted ones, the tracking module assumes that this is the same object and assigns corresponding ID. This detector has much better results at complex object tracking but still fails, when objects get temporary occluded with other similar objects, disappear and reappear in later frames.

With deep learning technology, it becomes possible to track objects almost as accurate as live brains. The key is the object’s appearance. Similar to humans, artificial neural network extract object’s appearance features, such as people’s clothing or car’s color and model. These features let the AI object tracking module compare the similarity of two objects in two subsequent frames and decide if this is the same object or a different. The deep learning object tracking module is robust and can deal with crowd-like scenes. It can keep tracking numerous objects at the same time even if they disappear for a quite long time from the frame and reappear far from the initial point.

Object tracking software configuration

BitRefine Heads recognition platform offers a number of ready-to-use modules that users can add to his image processing pipeline. We offer all three objects tracking modules, starting from the simplest one to the most advanced deep learning tracker. The user is free to decide which one he needs for his task. The main tradeoff is recognition speed. The more complex technology user includes into his processing pipeline, the more CPU/GPU power it requires.

Related Solutions

We’ve got more video analytics to share. Let’s be friends. But you first...