OpenMV Self-Training Neural Network for Target Point Detection—Traffic Sign Recognition in Real-Road Environments

Artificial intelligence has two very common tasks in the field of computer vision: classification and object detection.

Category:Determine which category the image belongs to—for example, whether it is a cat or a dog.

Object Detection:Used to locate the positions, quantities, and dimensions of multiple distinct objects in an image.

In our previous video tutorial, we explained how to use the OpenMV4 Plus to train neural networks online for classifying different objects. It can determine in real time whether an image within the OpenMV’s field of view contains a specific object—for example, whether a face is wearing a mask—but it cannot output the coordinates of faces or masks, nor the number of faces.

Target detection is significantly more complex and computationally intensive than classification, making it rare in the microcontroller domain. The main MCU of OpenMV is the STM32H7; today, we have added a new feature to OpenMV that enables target detection–like tasks to run on microcontrollers, delivering outstanding performance.

Today, we will explain to youTraining demonstrates traffic sign detection in real-road environments, detecting traffic signs on our actual roadssuch as no honking, no parking, and a speed limit of 80 km/h, rather than a laboratory environment free from interference.

You can view the entire project content, including all code and models, at the following address: https://book.openmv.cc/project/traffic-sign.html

It is widely recognized that real-world environments—especially road environments—are highly complex and contain numerous interfering factors. We captured images of traffic signs in actual environments using devices such as smartphones, trained the model on EdgeImpulse, the online platform of our OpenMV partner, and manually annotated approximately 270 images. The training process took only ten minutes, yielding a model with an F1 score accuracy of 92%. When deployed on OpenMV, the model achieves a frame rate of 10 frames per second, delivering both smooth performance and high accuracy.

Users may train SingTown to detect any target of interest—such as various numerals, different fruits, distinct markers, diverse components, or even any specific irregular object—by following our video tutorials, thereby enabling detection of the target’s count, coordinates, and object type name.

Note:This video tutorial covers the neural network-based keypoint detection feature, which is compatible not only with the OpenMV4 H7 Plus but also with the OpenMV4 H7. Since the trained FOMO keypoint detection model is compact—excelling in performance while occupying only several tens of kilobytes—even the OpenMV4 H7, which has less RAM than the Plus version, can run a slightly smaller variant of the model.

Below is a brief introduction to the principle:

The primary design decision behind FOMO object detection on OpenMV is based on the idea that many object detection tasks do not actually require the size of objects, but only their positions within an image. Once the positions of objects are known, subsequent operations can be performed—such as using OpenMV to control a robot car to move toward a detected target, enabling a drone to land at a specific location, or controlling a robotic arm to grasp a particular target object.

The FOMO model is a model specially designed by EdgeImpulse for microcontroller environments; unlike conventional object detection, it can detect only the positions and quantities of multiple objects, not their specific dimensions.

The underlying principle of the FOMO model is very simple and flexible. It first divides the image into blocks, each measuring 8×8 pixels. For an image with a resolution of 96×96, this results in a 12×12 grid of cells. For an image with a resolution of 360×360, this results in a 40×40 grid of cells. Then, image classification is performed independently on each cell.

FOMO performs significantly better than YOLO V5 or MobileNet SSD on a large number of small objects. FOMO object point detection is 30 times faster than MobileNet SSD.

FOMO performs better when the objects to be detected are of similar sizes—for instance, when the markers you aim to identify are roughly uniform in size, rather than varying greatly between very large and very small objects.

OpenMV uses EdgeImpulse to train neural network object detection models online, primarily involving the following steps: collecting and uploading image datasets, labeling objects, training, testing the model, and deployment.

* Collection of Image Datasets: Capture images of specific objects in the actual environment where recognition will be performed. Images are collected in the same environment where recognition will occur, ensuring that the actual training environment closely matches the actual detection environment, thereby achieving better results.

* Annotation Target: Draw bounding boxes around each target to be detected in our dataset and label them with the corresponding target names to facilitate subsequent training, detection of target locations, and identification of target names.

* Training the Model: After annotation is completed, train the model using the convolutional neural network parameters designed by us.

* Test Model: We conduct testing using the trained model; if the results are unsatisfactory, we correspondingly expand the training dataset or adjust the training parameters and continue training until we obtain a satisfactory model.

*Deployment: We simply place the trained model file onto the OpenMV’s built-in USB drive and run it.

Below is a demonstration of the specific process:

01. Collect and upload the image dataset

I. First, log in to the website edgeimpulse.com, select “Log In,” enter the name of your new project, and choose “Images” → “Classify multiple objects”

II. Select an image to upload. There are several ways to obtain the image dataset:

1. Capture images in real time using OpenMV IDE. Use the “Dataset Editor” tool in OpenMV IDE to create a new dataset and capture images in real time; for details, refer to our previous video tutorial on object classification for mask detection.

2. Use images downloaded from the internet or photos taken with a mobile phone. Ideally, these should match your actual inspection environment. We have prepared approximately four to five hundred images of traffic signs, including “No Parking,” “No Honking,” and “Speed Limit 80.” Each image was captured on real roads. This dataset is available for download from our GitHub repository and tutorial website.

02. Labeling Target

Upload approximately 100 images for each category; after uploading the images, proceed to annotate them by marking both the location and category of the objects to be identified.

03. Train the Model

Configure the training parameters, change the resolution to 128×128, and select all default parameter settings; then click Save.

Generate features, with three colors representing three distinct markers.

Configure the target detection training parameters. The default number of training epochs is 60, the learning rate is 0.001, and the validation set ratio is 20%. Enable data augmentation. For transfer learning, select either the default MobileNetV2 0.35 or MobileNetV2 0.1. (Note: SSD cannot be used, as it is not supported on OpenMV.)

Select Start Training. Upon completion of training, the neural network model achieves an F1 Score of 91.2%, and we can save the current version.

Select Storage, enter your description, then select Save.

04. Deploy the Model

Export the trained model file. Only export the library; select OpenMV and click Build to begin deploying and exporting the model.

It will automatically download the exported model, which in this case is the model trained on our 300 images, thereby completing the entire training process.

05. Run the Model

Run the model on OpenMV:

Connect the OpenMV Plus, save the three trained files to the OpenMV’s built-in USB drive, and open the ei_object_detection.py file in the OpenMV IDE. Click Run, and the results will appear in the serial terminal.

This feature is also compatible with the OpenMV4. Since the OpenMV4 does not have external SDRAM and has less memory, we can reduce the model size to enable it to run on the OpenMV4.

There are two methods to reduce the model’s size:

I. Reduce the training resolution from 128×128 to 96×96.

II. Modify the transfer learning model used by changing MobileNetV2 0.35 to MobileNetV2 0.1.

Finally, place the trained new file into the OpenMV4 and run it.

The above is our target point detection tutorial. We look forward to your results!