Object Tracking
|
Requires a Blickfeld QbProtect. |
The continuous monitoring and analysis capabilities of the QbProtect are greatly enhanced by detecting and tracking objects throughout the scene. Specialized zones (e.g., object-based security zone) used in advanced security setups require information derived from object tracking to generate alarms with a false alarm rate close to zero.
Object tracking is split into three consecutive steps:
- Detection
-
The foreground point cloud is clustered and divided into smaller, cohesive segments. The resulting list of detected objects is the basis for object tracking.
- Tracking
-
Object tracking ensures consistent association of individual detected objects over time. This is achieved by assigning a unique id to each object. The derived attributes (e.g., lifetime, velocity, etc.) attached to each object are then used by specialized zones to generate alarms.
- Classification
-
Detected objects are classified either by size or by a pre-trained model.
-
If no model is specified, clusters are categorized into predefined object size categories.
-
If a model is provided, classification is performed after the tracking step using the model, which assigns specific labels and confidence score to the tracked objects.
-
|
The main difference between detected and tracked objects is that detected objects are identified in a single frame, while tracked objects are continuously followed across multiple frames to maintain their identity and motion. |
|
The list of tracked objects generated in the tracking step can also be requested as a standalone data stream via the sensor’s API (e.g., for visualization). |