hailo-apps / rules

---
description: "Skill: Camera Setup & Management"
alwaysApply: false
---

# Skill: Camera Setup & Management

> Configure and manage camera sources for Hailo applications.

## When to Use This Skill

- User needs to set up a **USB camera**, **RPi camera**, or **RTSP stream**
- User wants to **discover available cameras** on the system
- User needs camera **configuration** (resolution, FPS, format)

## Camera Type Detection

```python
from hailo_apps.python.core.gstreamer.gstreamer_helper_pipelines import get_source_type
from hailo_apps.python.core.common.defines import USB_CAMERA, RPI_NAME_I

source_type = get_source_type(video_source)
# Returns: "usb", "rpi", "file", "rtsp", "ximage"
```

## USB Camera Discovery

```python
from hailo_apps.python.core.common.camera_utils import get_usb_video_devices

devices = get_usb_video_devices()
# Returns: list of device indices, e.g., [0, 2]
# Uses udevadm to enumerate real cameras (filters out virtual devices)
```

## Camera Initialization Patterns

### USB Camera (OpenCV)
```python
import cv2

camera_index = get_usb_video_devices()[0]  # First available USB camera
cap = cv2.VideoCapture(camera_index)
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 480)
cap.set(cv2.CAP_PROP_FPS, 30)

# Read frames
ret, frame = cap.read()  # frame is BGR numpy array

# Cleanup
cap.release()
```

### RPi Camera (Picamera2)
```python
from picamera2 import Picamera2

picam2 = Picamera2()
config = picam2.create_preview_configuration(
    main={"size": (640, 480), "format": "RGB888"}
)
picam2.configure(config)
picam2.start()

# Read frames
frame = picam2.capture_array()  # frame is RGB numpy array (note: not BGR!)

# Cleanup
picam2.stop()
```

### RTSP Stream
```python
# Replace with your camera's actual RTSP URL and credentials
cap = cv2.VideoCapture("rtsp://<username>:<password>@<camera-ip>:554/stream")
```

## CLI Source Selection

The standard parser handles source selection:
```bash
# USB camera
python3 my_app.py --input usb

# RPi camera
python3 my_app.py --input rpi

# Video file
python3 my_app.py --input /path/to/video.mp4

# RTSP
python3 my_app.py --input rtsp://camera_ip:554/stream
```

## Camera Abstraction Pattern

For apps that need to support multiple camera types cleanly:

```python
def init_camera(source, source_type):
    """Initialize camera and return (get_frame, cleanup, name) tuple."""
    if source_type == "rpi":
        from picamera2 import Picamera2
        picam2 = Picamera2()
        config = picam2.create_preview_configuration(main={"size": (640, 480), "format": "RGB888"})
        picam2.configure(config)
        picam2.start()
        return (lambda: picam2.capture_array(), lambda: picam2.stop(), "RPi")
    else:
        cap = cv2.VideoCapture(source)
        cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
        cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 480)
        return (
            lambda: (lambda r: r[1] if r[0] else None)(cap.read()),
            lambda: cap.release(),
            "USB"
        )
```

## Important Notes

- USB cameras return **BGR** frames (OpenCV default)
- RPi cameras can return **RGB** frames (with `format="RGB888"`)
- Always convert to RGB before sending to VLM: `cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)`
- VLM expects **336×336** images — use `Backend.convert_resize_image()`
- Camera release is critical — always clean up in `finally` blocks