lib.sedna.core.multi_edge_inference.components.detector

Module Contents

Classes

ObjectDetector

In MultiEdgeInference, the Object Detection/Tracking component
class lib.sedna.core.multi_edge_inference.components.detector.ObjectDetector(consumer_topics=['enriched_object'], producer_topics=['object_detection'], plugins: List[sedna.core.multi_edge_inference.plugins.PluggableNetworkService] = [], models: List[sedna.core.multi_edge_inference.plugins.PluggableModel] = [], timeout=10, asynchronous=False)[source]

Bases: sedna.core.multi_edge_inference.components.BaseService, sedna.core.multi_edge_inference.components.FileOperations

In MultiEdgeInference, the Object Detection/Tracking component is deployed as a service at the edge and it used to detect or track objects (for example, pedestrians) and send the result to the cloud for further processing using Kafka or REST API.

Parameters:

  • consumer_topics (List) – A list of Kafka topics used to communicate with the Feature Extraction service (to receive data from it). This is accessed only if the Kafka backend is in use.

  • producer_topics (List) – A list of Kafka topics used to communicate with the Feature Extraction service (to send data to it). This is accessed only if the Kafka backend is in use.

  • plugins (List) – A list of PluggableNetworkService. It can be left empty as the ObjectDetector service is already preconfigured to connect to the correct network services.

  • models (List) – A list of PluggableModel. By passing a specific instance of the model, it is possible to customize the ObjectDetector to, for example, track different objects as long as the PluggableModel interface is respected.

  • timeout (int) – It sets a timeout condition to terminate the main fetch loop after the specified amount of seconds has passed since we received the last frame.

  • asynchronous (bool) – If True, the AI processing will be decoupled from the data acquisition step. If False, the processing will be sequential. In general, set it to True when ingesting a stream (e.g., RTSP) and to False when reading from disk (e.g., a video file).

Examples

model = ByteTracker() # A class implementing the PluggableModel abstract class (example in pedestrian_tracking/detector/model/bytetracker.py) objecttracking_service = ObjectDetector(models=[model], asynchronous=True)

Notes

For the parameters described above, only ‘models’ has to be defined, while for others the default value will work in most cases.

process_data(ai, data, **kwargs)[source]

The user needs to implement this function to call the main processing function of the AI model and decide what to do with the result.

preprocess(data)[source]

The user can override this function to inject some preprocessing operation to be executed before the data is added to the data structure by the ‘put()’ function.

close()[source]
update_operational_mode(status)[source]

The user needs to trigger updates to the AI model, if necessary.