Object tracking, how to get started
Before starting with Object Tracking and Object detection you must make sure that you have installed all the necessary libraries.
If you don’t have Opencv installed this is the command to run
pip install opencv-python
Download the files that we will use in this tutorial through the link at the bottom of the page.
1. Object detection
For convenience, I have already written this part and you find everything in the object_detection.py file. For this tutorial, we used Yolo v4 with the pre-trained model but if you need a custom model I suggest you see Train YOLO to detect a custom object (online with free GPU).
To use it just a call in the main file
...
from object_detection import ObjectDetection
...
# Initialize Object Detection
od = ObjectDetection()
while True:
...
# Detect objects on frame
(class_ids, scores, boxes) = od.detect(frame)
for box in boxes:
(x, y, w, h) = box
cx = int((x + x + w) / 2)
cy = int((y + y + h) / 2)
center_points_cur_frame.append((cx, cy))
...
Here is the result in the video frame.
2. Object Tracking
By saving the position of the center point of each object, you can trace the previous position of the objects and predict what the immediate next will be. Here is a small example in the image
Before going on with the explanation, I must point out that object tracking is more complicated than that. Consider this tutorial as a simple exercise to understand the basics behind this algorithm.
For practical use I suggest you use SORT and Deep SORT industrial applications, you will find concrete examples and complete codes on my course dedicated to object detection and object tracking.
Find the point and assign the ID
We don’t need the history of all the tracking but only the last points so Initialize an array to keep track of the previous points and then we need to calculate the distance between the points to make sure they all belong to the same object. The closer the points are, the greater the probability that we are tracking the same object.
...
# Initialize count
count = 0
center_points_prev_frame = []
tracking_objects = {}
track_id = 0
...
# Only at the beginning we compare previous and current frame
if count <= 2:
for pt in center_points_cur_frame:
for pt2 in center_points_prev_frame:
distance = math.hypot(pt2[0] - pt[0], pt2[1] - pt[1])
if distance < 20:
tracking_objects[track_id] = pt
track_id += 1
...
As you can see from the portion of code above with the math.hypot() function the distance of the two points is calculated and if the distance is less than 20, an ID is associated with the position of the point.
Assign univocal ID
In this part of the code, we have to make sure to compare the previous object with the current one and update the position of the ID. In this way, the same object remains with the same ID for its entire path. When the object is no longer recognized, it loses the ID.
...
else:
tracking_objects_copy = tracking_objects.copy()
for object_id, pt2 in tracking_objects_copy.items():
object_exists = False
for pt in center_points_cur_frame_copy:
distance = math.hypot(pt2[0] - pt[0], pt2[1] - pt[1])
# Update IDs position
if distance < 20:
tracking_objects[object_id] = pt
object_exists = True
continue
# Remove IDs lost
if not object_exists:
tracking_objects.pop(object_id)
...
Now the tracking works quite well and as you can see from the image below, the white car has lost track because the object has not been identified anymore.
Add new ID to new cars
If a new object is identified, the list of points must also be updated. So here are the changes of the previous code that allow you to delete the old and add the points of the new cars identified.
...
else:
tracking_objects_copy = tracking_objects.copy()
center_points_cur_frame_copy = center_points_cur_frame.copy()
for object_id, pt2 in tracking_objects_copy.items():
object_exists = False
for pt in center_points_cur_frame_copy:
distance = math.hypot(pt2[0] - pt[0], pt2[1] - pt[1])
# Update IDs position
if distance < 20:
tracking_objects[object_id] = pt
object_exists = True
if pt in center_points_cur_frame:
center_points_cur_frame.remove(pt)
continue
# Remove IDs lost
if not object_exists:
tracking_objects.pop(object_id)
# Add new IDs found
for pt in center_points_cur_frame:
tracking_objects[track_id] = pt
track_id += 1
...
As can be seen from the image below, the id is kept until the object is recognized. In fact, in the central part of the video, it traces very well.
3. Object Tracking for industrial use
I made this tutorial to explain the basics of object tracking and I invite you to practice creating simple tracking codes. If you want something a little more advanced than this, look at Object Tracking with Opencv and Python but if you need use in real conditions I recommend my Build COMPUTER VISION PROJECTS course which also contains Object tracking with Deep SORT.
Hi there, I’m the founder of Pysource.
I’m a Computer Vision Consultant, developer and Course instructor.
I help Companies and Developers to build efficient computer vision software.