Edge AI with Intel Neural Compute Stick and ASI Biont: On-Device ML Integration Guide

The Intel Neural Compute Stick 2 (NCS2) is a USB-based AI accelerator that brings deep learning inference to edge devices. When paired with ASI Biont, an AI agent that automates device integration, you can unlock powerful on-device ML capabilities without the need for cloud APIs. This article provides a practical guide to connecting the NCS2 to ASI Biont, running local inference with OpenVINO, and automating video analytics and sensor data processing. We'll cover the connection method, code examples, and how to set it all up through the AI-powered integration builder on asibiont.com.

Why Integrate Intel Neural Compute Stick with ASI Biont?

In the era of IoT, latency and data privacy are critical. Offloading AI inference to the cloud introduces delays and security risks. The Intel Neural Compute Stick 2, with its Myriad X VPU, enables on-device inference, but managing the device and integrating it with your automation workflows can be complex. ASI Biont simplifies this by providing a natural language interface to connect and control the NCS2. You can describe what you want in plain English, and ASI Biont writes the Python code to make it happen. This integration reduces costs (no cloud fees), lowers latency (millisecond responses), and keeps sensitive data local.

Understanding the Intel Neural Compute Stick 2

The Intel Neural Compute Stick 2 (NCS2) is a USB 3.0 device that uses the Intel Movidius Myriad X Vision Processing Unit (VPU). It supports 4 TOPS of compute performance and is designed for deep learning inference at the edge. It works with the Intel OpenVINO toolkit, which optimizes and runs models in intermediate representation (IR) format. Common use cases include object detection, facial recognition, and gesture recognition. The NCS2 is ideal for prototyping and low-power edge AI applications.

How ASI Biont Connects to the Device

ASI Biont is an AI agent that can connect to virtually any device through a conversational interface. For the Intel Neural Compute Stick, the primary connection method is via Python scripting using the execute_python universal connector. This allows ASI Biont to run code that leverages the OpenVINO runtime to perform inference. The user simply describes the task, and ASI Biont generates the necessary Python code, which is executed in a sandboxed environment. This approach eliminates the need for complex configuration panels or manual coding.

Connection Methods Overview

For the NCS2, we use execute_python because it allows direct interaction with the OpenVINO runtime, which is essential for loading models and running inference. The AI agent writes code that initializes the model, processes input data (e.g., camera frames), and outputs results.

Use Case: Real-Time Video Analytics on Edge

Imagine a security camera connected to a Raspberry Pi with an NCS2. The goal is to detect persons and alert via Telegram if an intruder is spotted. ASI Biont can automate this by:
1. Generating a Python script that captures video from a USB camera.
2. Running object detection using a pre-trained MobileNet SSD model on the NCS2.
3. When a person is detected, sending an alert via Telegram API.

Step-by-Step Integration

Step 1: Describe the Task in ASI Biont Chat

You type: "Connect to my Intel Neural Compute Stick via execute_python. Use OpenVINO to run MobileNet SSD for person detection. Capture video from /dev/video0, and if a person is detected, send a message to my Telegram bot."

Step 2: ASI Biont Generates the Code

ASI Biont writes a Python script using the OpenVINO runtime. Here's a simplified example:

import cv2
import numpy as np
from openvino.inference_engine import IECore
import requests

# Initialize OpenVINO
ie = IECore()
net = ie.read_network(model='mobilenet-ssd.xml', weights='mobilenet-ssd.bin')
exec_net = ie.load_network(network=net, device_name='MYRIAD')

# Capture video
cap = cv2.VideoCapture(0)

while True:
    ret, frame = cap.read()
    if not ret:
        break

    # Preprocess frame
    resized = cv2.resize(frame, (300, 300))
    input_blob = np.transpose(resized, (2, 0, 1))
    input_blob = input_blob.reshape(1, 3, 300, 300)

    # Run inference
    outputs = exec_net.infer(inputs={'input': input_blob})
    detections = outputs['detection_out']

    # Parse detections
    for detection in detections[0][0]:
        confidence = detection[2]
        if confidence > 0.5 and int(detection[1]) == 15:  # class 15 is person in COCO
            # Send Telegram alert
            requests.post('https://api.telegram.org/bot<YOUR_BOT_TOKEN>/sendMessage',
                          json={'chat_id': '<CHAT_ID>', 'text': 'Person detected!'})
            break

cap.release()

Note: The above code runs indefinitely; however, ASI Biont's execute_python has a 30-second timeout, so for continuous monitoring, you'd need to use a different approach, such as running the script via SSH or using a scheduled task. But for demonstration, ASI Biont can run a single inference on a static image or a short video segment.

Step 3: ASI Biont Executes the Code

ASI Biont runs the script in its sandbox environment. If the NCS2 is connected to the same machine, it will perform inference. The output can be logged or used to trigger further automation.

Alternative: Using Hardware Bridge for Serial Devices

If your device communicates via RS-232/RS-485, you can use the Hardware Bridge. For the NCS2, this is not applicable, but it's good to know for other devices. The bridge is downloaded from the ASI Biont dashboard and run locally:

python bridge.py --token=YOUR_TOKEN --ports=COM3 --baud 115200 --rate=10

Then, you can send commands via industrial_command in the chat. For example, to read a sensor value:

industrial_command(protocol='modbus', command='read_holding_registers', address=1, count=2)

Scenarios for Smart Home and Surveillance

With the NCS2 and ASI Biont, you can automate various tasks:

• Person Detection for Security: Detect intruders and send alerts.
• Pet Monitoring: Use a model to detect if your pet is on the sofa and notify you.
• Gesture Control: Recognize hand gestures to control smart home devices.
• Anomaly Detection in Industrial Sensors: Analyze sensor data patterns locally.

These scenarios benefit from low latency and privacy, as all data stays on-premises.

The Power of Universal Python Integration

ASI Biont's execute_python connector is a game-changer. You don't need to wait for specific device support—the AI agent writes custom Python code for any device. For example, to connect to a temperature sensor via Modbus, you can say: "Read temperature from 192.168.1.100 using Modbus, and if it exceeds 80°C, send an alert." ASI Biont will generate code using pymodbus and handle the logic. This flexibility means you can integrate the NCS2, Arduino, PLCs, and more, all through simple chat commands.

Why This Benefits You

Traditional integration requires deep programming knowledge and time. ASI Biont reduces this to seconds. You describe the task, and the AI writes, debugs, and executes the code. This accelerates prototyping and reduces the barrier to entry for edge AI. Moreover, by keeping inference on the edge, you save on cloud costs and ensure data privacy.

Getting Started with ASI Biont

To start integrating your Intel Neural Compute Stick with ASI Biont:

1. Sign up at asibiont.com.
2. In the chat, describe your device and task. For example: "I have an Intel Neural Compute Stick 2. I want to run a face detection model and send alerts when a face is recognized."
3. ASI Biont will ask for specifics (model path, camera index, alert channel) and then generate the code.
4. Run the code either in the sandbox or on your local machine via SSH.

Conclusion

The Intel Neural Compute Stick 2, when combined with ASI Biont, offers a powerful yet accessible edge AI solution. You can automate complex computer vision tasks without writing code from scratch. The integration is seamless, cost-effective, and privacy-preserving. Try it today at asibiont.com and see how easy it is to bring AI to the edge.