Skip to main content

Basic Application with Tracking

This example demonstrates the simplest way to build a complete inference application with the Voyager SDK. It runs a YOLOv5 tracker on two video streams, displays the results in a window, and prints vehicle tracking information to the terminal.

Use this as your starting point for any new application that needs object detection and tracking with visual output.

What you'll learn​

  • How to create an inference stream with multiple video sources
  • How to iterate over frame results and access tracker metadata
  • How to use the display module to show annotated frames in a window
  • How to extract bounding box history from tracked objects

Prerequisites​

  • Voyager SDK installed and activated
  • Sample media files available in media/ (included with the SDK)

Source​

Running this example

This example is included in the SDK at examples/application.py. From the SDK directory with your environment activated:

source venv/bin/activate
python examples/application.py
application.py
#!/usr/bin/env python
# Copyright Axelera AI, 2025
from axelera.app import config, create_inference_stream, display

stream = create_inference_stream(
    network="yolov5m-v7-coco-tracker",
    sources=[
        config.env.framework / "media/traffic1_1080p.mp4",
        config.env.framework / "media/traffic2_1080p.mp4",
    ],
)


def main(window, stream):
    window.options(0, title="Traffic 1")
    window.options(1, title="Traffic 2")
    VEHICLE = ('car', 'truck', 'motorcycle')
    center = lambda box: ((box[0] + box[2]) // 2, (box[1] + box[3]) // 2)
    for frame_result in stream:
        window.show(frame_result.image, frame_result.meta, frame_result.stream_id)
        for veh in frame_result.pedestrian_and_vehicle_tracker:
            print(
                f"{veh.label.name} {veh.track_id}: {center(veh.history[0])} → {center(veh.history[-1])} @ stream {frame_result.stream_id}"
            )

        if window.is_closed:
            break


with display.App(renderer=True) as app:
    wnd = app.create_window("Business logic demo", (900, 600))
    app.start_thread(main, (wnd, stream), name='InferenceThread')
    app.run()
stream.stop()

Key concepts​

The create_inference_stream function is the main entry point for building inference applications. It accepts a network name (here yolov5m-v7-coco-tracker) and a list of video sources. The SDK handles model compilation, pipeline setup, and frame decoding automatically.

Each frame_result yielded by the stream contains the annotated image, metadata, and a stream_id indicating which source produced the frame. The pedestrian_and_vehicle_tracker attribute provides tracked detections, each with a track_id and a history of bounding boxes that lets you trace an object's movement over time.

The display module follows a thread-based pattern: the App context manager owns the GUI event loop, while the inference logic runs in a separate thread started via app.start_thread. The window.show() call routes each frame to the correct panel based on its stream_id, and window.is_closed provides a clean shutdown signal.