Artificial Intelligence and 3D Vision for Fruit Sorting

Agricultural products can vary greatly in shape and color, which poses a great challenge for traditional image processing methods. One example is the recognition and sorting of fruit.

To master this task, Basler has teamed with the software provider Data Spree to develop a vision solution for the detection and classification of individual fruits, using 3D data as raw material. The spatial information gained during the teaching of the neural networks creates a robust real-time solution for fruit sorting.

Here's How It Works

The deep-learning-based vision system consists of a Basler blaze time-of-flight camera and a standard PC. The Basler blaze camera provides high-resolution 3D images with nearly millimeter precision. It not only generates a grayscale image as an intensity image, but also measures the distance to each individual pixel by measuring the travel time of light pulses in the near-infrared range. The resulting image is then available as a 3D point cloud and thus provides additional information about the imaged scene. As compared to 2D RGB images, in this approach the color information is replaced by shape information, which not only has advantages in the simultaneous detection of red and green apples, but also enables additional applications, such as the exact positioning and measurement of the detected objects.

Discover our 3D-vision portfolio

3D point cloud image of the fruits

3D Image Processing and Artificial Intelligence (AI) as a Strong Double

The software is an essential part of this application. It consists of two main components: the Basler blaze SDK and the AI application software.

The user-friendly and platform-independent programming interface of the Basler blaze allows easy integration of the Data Spree software "Deep Learning DS". This software solution, based on deep neural networks (deep learning), is extremely user-friendly and allows the development of deep learning models without any previous knowledge. With the help of the application software, the individual work steps for system design, such as data acquisition, annotation, training, provisioning and implementation of the trained network on the target hardware can be significantly simplified. Learn more about the individual work steps in the Use Case.

Download the complete Use Case now

Training of the neural network

The Advantages of this 3D Vision Solution with Basler blaze at a Glance

Basler offers not only 3D vision hardware, but also support in implementing the corresponding software solutions. If required, we also activate our extensive partner network to achieve customer requirements.

User-friendly, platform-independent programming interface (Basler blaze SDK) with sample programs
Use of industry-proven and durable camera hardware with IP67 protection class
Easy hardware installation due to integrated illumination and calibrated optics
More precise and reliable recognition and classification of objects by integrating the spatial information via a 3D camera (time-of-flight) into the learning of the neural networks
Reduced complexity of the application, as complementary sensor technology is no longer necessary in a large number of applications
Precise measurement results even in low light, daylight and without contrast

More about the Basler blaze camera

Do You Have Any Questions? We are Happy to Help.

Do you have any questions about this solution, your individual solution or our products? Please contact our Sales Team directly.

Contact the Sales Team