Implementing Personal Detection System with OpenVINO: Add Audio Alerts

Table of Contents

1. Introduction
2. Requirements
3. Installation
4. Setting Up the Environment
5. Downloading the Code and Models
6. Preparing the Input Data
7. Loading the Model
8. Pre-processing the Image
9. Performing Inference
10. Visualizing the Results
11. Adding Audio Alerts

Introduction

In this article, we will explore how to implement a personal detection system using OpenVINO models. We will cover the installation process, setting up the environment, downloading the necessary code and models, preparing the input data, loading the model, performing inference, visualizing the results, and adding audio alerts.

Requirements

Before getting started, make sure you have the following requirements:

• OpenVINO Package installed
• Access to a compatible hardware device
• Basic knowledge of Python and computer vision

Installation

To install the OpenVINO package, refer to the official documentation provided by Intel. The installation process may vary depending on your operating system (Linux, Windows, Mac OS, or Raspbian).

Setting Up the Environment

Assuming you have installed and activated your OpenVINO environment, you need to download the code and models from the GitHub repository. Additionally, you can download a sample video (in MP4 format) or use your web camera as the input source.

Downloading the Code and Models

Once you have downloaded the necessary files, extract them. Create a folder named "model" and another folder named "video". Copy the person_media.mp4 file into the "video" folder. Place the pedestrian detection models (adas_002.xml and adas_002.bin) inside the "model" folder.

Preparing the Input Data

In your chosen code editor (such as VS Code), open the main file of the person detection code. Set the necessary variables, such as the model XML and model bin files. Define the input stream (e.g., the person_media.mp4 video or webcam).

Loading the Model

Using the Inference Engine (IE) Core, load the model by providing the path to the model XML file. Specify the target device (e.g., CPU), and create a network object. Then, load the network with the specified device and number of request scores.

Pre-processing the Image

Pre-process the input image by resizing it to the required Dimensions (batch size, Channel, Height, and width). Capture the frames from the input stream using OpenCV, and preprocess them accordingly. Set the initial request ID to 0.

Performing Inference

Start an infinite loop to process the frames. Measure the start time to calculate the frames per Second (FPS). Resize the frame, transpose it, and reshape it to match the required dimensions. Start the inference cycle. Once the inference is complete, measure the end time and calculate the FPS value. Retrieve the detection results.

Visualizing the Results

Iterate through the results and extract the probability and box coordinates for each detected object. Apply a threshold to filter out low probability detections. Create bounding boxes and Visualize them on the frame. Display the FPS value and other Relevant information on the frame.

Adding Audio Alerts

To add audio alerts, import the necessary libraries (such as datetime, timedelta, and pygame). Initialize the Pygame mixer and create an alert time list. Within the alert function, use an if statement to determine when to play the audio file based on the interval. Append new alert times to the list.

Conclusion

Congratulations! You have successfully implemented a personal detection system using OpenVINO models. The system is capable of performing inference on input data, visualizing the results, and providing audio alerts for detected intruders. Experiment with different settings and explore further enhancements to improve the functionality and performance of the system.

Resources

• OpenVINO: official documentation