Detection of Pedestrians in Reverse Camera Using Multimodal Convolutional Neural Networks

Area of Science:

• Computer Science
• Artificial Intelligence
• Automotive Engineering

Background:

• Road safety is a global concern, with pedestrian detection systems crucial for reducing accidents.
• Current systems often focus on forward detection, neglecting risks associated with vehicles reversing.
• Pedestrian collisions during reverse driving pose a significant safety hazard.

Purpose of the Study:

• To develop and evaluate an AI model for detecting pedestrians behind vehicles during reverse driving.
• To fuse data from backup cameras and ultrasonic sensors for enhanced detection accuracy.
• To contribute to the development of intelligent automotive safety systems.

Main Methods:

• A novel model combining a one-dimensional convolutional neural network (CNN) and the Inception V3 architecture was proposed.
• Information from vehicle backup cameras and ultrasonic sensors was fused.
• A dedicated database was created through specific data collection for training and validation.

Main Results:

• The proposed CNN model achieved high performance in pedestrian detection.
• The model demonstrated 99.85% accuracy and 99.86% correct classification.
• The fusion of camera and sensor data proved effective for robust detection.

Conclusions:

• The research successfully demonstrated the efficacy of CNNs for detecting pedestrians during reverse driving.
• Fusing data from multiple sensors significantly improves the reliability of pedestrian detection systems.
• The developed model offers a viable solution for enhancing automotive safety and preventing accidents.