Enhanced Road Safety: AI-Powered Radar Systems for Pedestrian and Cyclist Detection


CoE-EDP, VisionRICoE-EDP, VisionRI | Updated: 14-06-2024 16:00 IST | Created: 14-06-2024 16:00 IST
Enhanced Road Safety: AI-Powered Radar Systems for Pedestrian and Cyclist Detection
Representative Images

Researchers Blazej Slesicki and Anna Slesicka from the Faculty of Aviation Division, Polish Air Force University, and Institute of Navigation, Polish Air Force University, Poland explore an innovative method to improve road safety by utilizing frequency-modulated continuous-wave (FMCW) radar combined with deep learning algorithms. The study focuses on addressing the increasing number of accidents involving pedestrians and cyclists, emphasizing the potential of advanced radar systems in autonomous vehicles to enhance safety.

Innovative Use of FMCW Radar and Deep Learning

The research involves using FMCW radar to capture radar signatures of various objects, including pedestrians, cyclists, and cars. These signatures are then processed using a convolutional neural network (CNN) specifically designed for this task. Radar data is generated in a controlled Matlab environment, creating a database of radar signatures that CNN uses to learn and distinguish between different types of objects. One of the significant contributions of this work is the novel method for discriminating between multiple objects within a single radar signature. This method leverages the short-time Fourier transform (STFT) to convert radar data into spectrograms, which the CNN then analyzes to identify and classify objects.

Detailed System Architecture and Performance

The study presents a detailed explanation of the radar system and the CNN architecture. The radar system operates at 77 GHz, providing high-resolution data suitable for object detection and classification. The CNN consists of multiple layers designed to process and interpret the radar data accurately. The network includes convolutional layers, batch normalization layers, and max-pooling layers, culminating in a final classifier that distinguishes between ten different classes of objects. The experiments conducted in the study demonstrate the high accuracy of the proposed system. The CNN was trained and tested on a dataset containing 10,000 images, achieving an accuracy of 95.93% in recognizing different traffic participants. The study compares the performance of the proposed method with other existing methods, showing superior results due to the carefully designed network structure and the use of simulated radar data.

Extensive Practical Applications

The practical applications of this research are extensive. The proposed system can be integrated into intelligent vehicle systems to provide real-time warnings to drivers, enhancing their ability to detect and avoid collisions with pedestrians, cyclists, and other vehicles. The system can also be adapted for use in various sectors, such as production halls, where it can prevent accidents by automatically detecting and classifying moving objects. The research underscores the importance of deep neural networks and deep learning in advancing artificial intelligence applications. By combining these technologies with radar, the study opens up possibilities for real-world applications that can benefit everyday life.

Leveraging Deep Neural Networks for Safety

Deep neural networks, which generate diagnostic features and act as classifiers, can process original data without preliminary expert analysis. This efficiency is particularly beneficial in the context of radar data, where accurate interpretation is crucial for safety applications. The authors propose a missing approach based on simulated data generated in the Matlab environment and a novel method of identifying objects using an FMCW radar operating at 77 GHz. The radar signals are processed using STFT, and the resulting spectrograms are analyzed with a specialized deep neural network structure. The system's ability to distinguish between multiple objects in a single radar signature marks a significant advancement over existing techniques.

Future Directions and Practical Implementation

The proposed approach demonstrates the potential for creating intelligent vehicles that provide real-time pedestrian detection and warning mechanisms. By vibrating the steering wheel and displaying messages on the dashboard, the system can alert drivers to potential collisions with pedestrians. Additionally, the solution can be used in automatic detection systems for people, animals, or vehicles in various economic sectors, such as production halls, where it can prevent disruptions and ensure safety. The study's experimental results confirm the high efficiency of the developed method. In comparing the proposed method with other available methods, the authors implemented techniques described in related literature within the Matlab environment. The results show that the proposed method achieves the highest accuracy, attributed to the appropriate network structure and learning process for the input data.

The paper presents a comprehensive approach to traffic participant recognition using advanced radar technology and deep learning. The proposed system offers promising solutions for enhancing safety in autonomous and conventional vehicles. The obtained results of simulations and positive tests provide a basis for applying this system in various sectors and areas of the economy. Future research will focus on practical implementation, including acquiring necessary FMCW radar equipment, conducting practical tests, and refining the convolutional neural network structure.

  • FIRST PUBLISHED IN:
  • Devdiscourse
Give Feedback