Object Detection with OpenCV and PyTorch

A quick start to object detection with openCV + Pytorch + Kafka

Table of contents

1. Overview
2. Getting Started
3. Configurations
4. References

---

Overview

This is a demo project to show how to run inference with the default model as a YoloV5 ultralytics model.

• OpenCV to connect with a camera (/dev/video0) and save images to .build/capture
• Pytorch load model and run inference on images
• Kafka to provide a publisher/subscriber messaging fabric for clients

---

Getting Started

• first, lets' copy the env.example (view more below for how to customize it)

cp env.example .env

• now you are ready to spin it up

$ docker compose up
[+] Running 3/0
 ✔ Container kafka      Running                                                                                                                                                                                       0.0s 
 ✔ Container kafkacat   Running                                                                                                                                                                                       0.0s 
 ✔ Container inference  Running                                                                                                                                                                                       0.0s 
Attaching to inference, kafka, kafkacat
inference  | 11/30/2023 02:58:14 - thread:140222845712128 - publish.kafkaPub - INFO | Sending payload: {'event': 'no', 'datetime': '2023-11-30T02:58:14', 'alert_count': 0}
kafkacat   | % Reached end of topic test [0] at offset 110
inference  | 11/30/2023 02:58:17 - thread:140222845712128 - publish.kafkaPub - INFO | Sending payload: {'event': 'no', 'datetime': '2023-11-30T02:58:17', 'alert_count': 0}
kafkacat   | % Reached end of topic test [0] at offset 111

Notice above that kafkacat is used to receive messages. You can open a terminal and do some checks

# check the health of the broker, and list any topics
$ docker exec -it kafkacat kafkacat -b kafka:9092 -L
# subscribe to the topics
$ docker exec -it kafkacat kafkacat -b kafka:9092 -t test -C

If you want to see the detections, you can view them in .build/capture. The program will save images, cleaning itself up after every 100 images.

---

Configurations

inference app

• Check out env.example to see some variables that set up the application configs in src/config/config.py
• you should create an .env in src/config/.env so that the application can load it

# model
THRESHOLD=0.6                               // threshold for object detection
TORCH_MODEL_DIR='ultralytics/yolov5'        // model = torch.hub.load(TORCH_MODEL_DIR, TORCH_MODEL, pretrained=True)
TORCH_MODEL='yolov5s'
CLASSES=person                              // the single class label that you want to detect from the model

# camera
CAPTURE_DELAY=1                             // capture a frame every X seconds. FPS=1/CAPTURE_DELAY  
CAM_WIDTH=640                               // resolution from /dev/video0
CAM_HEIGHT=480

# kafka producer                            // path to the json file that describes the server and topic
KAFKA_CONFIGS=/application/src/publish/config.json

kafka server

• Before doing a build, navigate to .build and see three files that set up the kafka server. Click the links below for examples

  • kafka: server.properties Sets properties for the kafka server
  • zookeeper: zookeeper.properties Sets properties for the zookeper broker within the kafka server
  • supervisorD: supervisord.conf Runs a controlled process in the docker container.

• Notice that src/publish/config.json has a field to publish results to the kafka server at kafka:9092 to a topic named test

You do not need to change any of the configurations above, this is just for reference.

---

References

update docker-compose

• this is for linux machines!

# check that version is 1.29 or greater
$ docker-compose --version

# remove old version
$ sudo rm /usr/local/bin/docker-compose

# download new version (adjust to a version that you'd like)
$ export COMPOSE_VERSION=1.29.2
$ sudo curl -L "https://github.com/docker/compose/releases/download/${COMPOSE_VERSION}/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
$ sudo chmod +x /usr/local/bin/docker-compose

# test 
$ docker-compose --version
docker-compose version 1.29.2, build 5becea4c

---