Object classification in automotive ultrasonic surround sensing using a compact 2 × 2 sensor array and a...

Area of Science:

• Automotive Engineering
• Sensor Technology
• Machine Learning

Background:

• Current park pilot systems rely on ultrasonic surround sensing, limited by single-element sensor performance.
• Object classification is critical for advanced driver assist systems but hindered by the lack of directional data in conventional ultrasonic sensors.

Purpose of the Study:

• To enhance object classification accuracy in ultrasonic surround sensing.
• To evaluate the effectiveness of a small 2x2 array ultrasonic sensor compared to single-element sensors.

Main Methods:

• Proposed replacing single-element ultrasonic sensors with a 2x2 array sensor.
• Developed an efficient convolutional neural network (CNN) for classifying preprocessed transducer signals.
• Evaluated feature extraction methods including beamforming and acoustic source maps.

Main Results:

• The 2x2 array sensor, combined with CNN and acoustic source maps, achieved promising classification accuracies.
• The proposed array sensor system significantly outperformed conventional single-element sensors in object classification.
• Incorporating directional information from the array sensor enhanced target discrimination.

Conclusions:

• Small aperture array sensors offer a significant improvement for object classification in ultrasonic surround sensing.
• Leveraging directional information is key to advancing the performance of ultrasonic perception systems.
• The developed CNN approach effectively classifies signals from the array sensor, demonstrating practical application potential.