One of the basic task in computer vision is object counting. AI Singapore developed a simple solution built in conjunction with our object detection models. This can be used with CCTVs in malls, shops or factories for crowd control, or other general object counting. For advanced counting, check out zone counting.
Counting is done by looking at the count of objects detected by the object detection models. As an example, we can count the number of people that appear in a video, as per our gif above. This is explained in a subsequent section.
To try our solution on your own computer with PeekingDuck installed: use the following configuration file: object_counting.yml and run PeekingDuck.
> peekingduck run --config_path <path_to_object_counting.yml>
The main component to obtain the count is the detection from the object detection model, which is the bounding boxes.
1. Object Detection
We use an open source object detection estimation model known as Yolov4 and its smaller and faster variant known as Yolov4-tiny to identify the bounding boxes of chosen objects we want to detect. This allows the application to identify where objects are located within the video feed. The location is returned as two (x, y) coordinates in the form [x1, y1, x2, y2], where (x1, y1) is the top-left corner of the bounding box, and (x2, y2) is the bottom-right. These are used to form the bounding box of each object detected. For more information in how adjust the yolo node, checkout the Yolo configurable parameters.
We can also use the EfficientDet model as a more accurate but slower alternative.
2. Object Counting
To count the number of objects detected, we simply take the sum of the number of bounding boxes detected for the object.
These are the nodes used in the earlier demo (also in object_counting.yml):
nodes:
- input.live
- model.yolo:
- detect_ids: [0]
- dabble.bbox_count
- draw.bbox
- draw.bbox_count
- draw.fps
- output.screen
1. Object Detection Node
By default, the node uses the Yolov4-tiny model for object detection, set to detect people. To use more accurate models, you can try the Yolov4 model, or the EfficientDet model that is included in our repo.
2. Object Counting Node
The object counting node is called by including dabble.bbox_count in the run config declaration. This takes the detected bounding boxes and outputs the total count of bounding boxes. The node has no configurable parameters.
3. Adjusting Nodes
The object counting node does not have changeable configurations. However, it depends on the configuration set in the object detection models, such as the type of object to detect, etc. As such, please see the Yolo node documentation or the Efficientdet node documentation for adjustable behaviours that can influence the result of the object counting node.
For more adjustable node behaviours not listed here, check out the node glossary.
We have a more complex variant of object counting that is called zone counting which makes use of the zone count node. It allows for the creation of zones within a single image, and provides separate counts of the chosen objects detected for objects that fall inside the zones created.
For more information, check out the readme for zone counting.