Automatic brake Driver Assistance System based on deep learning and fuzzy logic

• Published in: PloS one 2024-12, Vol.19 (12), p.e0308858
• Main Authors: García-Escalante, A R, Fuentes-Aguilar, R Q, Palma-Zubia, A, Morales-Vargas, E
• Format: Article
• Language: English
• Subjects: Accidents, Traffic - prevention & control; Advanced driver assistance systems; Algorithms; Automation; Automobile Driving; Automobiles; Autonomous vehicles; Biology and Life Sciences; Brake pressing; Braking; Braking systems; Cameras; Classification; Computer and Information Sciences; Datasets; Decision making; Deep Learning; Detectors; Developmental stages; Embedded systems; Emergency preparedness; Emergency response; Engineering and Technology; Fuzzy algorithms; Fuzzy Logic; Fuzzy systems; Human error; Humans; Light; Mechanization; Physical Sciences; R&D; Research & development; Roads & highways; Safety and security measures; Sensors; Social Sciences; Synchronism; Synchronization; Traffic; Traffic signals
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0308858

Advanced Driver Assistance Systems (ADAS) aim to automate transportation fully. A key part of this automation includes tasks such as traffic light detection and automatic braking. While indoor experiments are prevalent due to computational demands and safety concerns, there is a pressing need for research and development of new features to achieve complete automation, addressing real-world implementation challenges by testing them in outdoor environments. These systems seek to provide precise synchronization for decision-making processes and explore algorithms beyond emergency responses, enabling braking actions with short reaction times. Therefore, this work proposes a level 1 ADAS for automatic braking. The implementation uses an NVIDIA Jetson TX2 and a ZED stereo camera for traffic light detection, which, in addition to the depth map provided by the camera and a fuzzy inference system, make the decision to perform automatic braking based on the distance and current state of the traffic light. The contributions of this research work are the development and validation of a one-stage traffic light state detector using EfficientDet D0, a brake profile using fuzzy logic, and the validation with an on-road experiment in Mexico. The traffic light detection model obtained a mAP of 0.96 for distances less than 13 m and 0.89 for 15 m, with an average RMSE of 0.9 m and 0.05 m in the braking force applied, respectively. Integrated systems have a response time of 0.23 s, taking a step further in the state-of-the-art.