Real-time obstacle detection by stereo vision and ultrasonic data fusion. (28th February 2022)

• Record Type: Journal Article
• Authors: Gholami, Farshad; Khanmirza, Esmaeel; Riahi, Mohammad
• Is Part Of: Measurement. Volume 190(2022)
• Publication Date: 2022-02-28
• Subjects: Stereo vision -- Ultrasonic -- Data fusion -- Intelligent vehicle; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.110718
• Languages: English
• ISSNs: 0263-2241

Abstract

Abstract: Advanced driving assistance systems include some features such as changing lanes, avoiding obstacles and moving in cruise mode. Creating the perception of the environment is one of the most critical steps in developing an advanced driving assistance system. One of the main tasks in the perceptual layer is the timely detection of obstacles. In this study, a mixed method underpinned by the ultrasonic sensors and stereo vision was proposed to reduce the computational complexity in achieving acceptable frequency in the perception of the environment. The modified cross-check is employed for quality improvement while maintaining real-time. Moreover, we proposed an implementation form for the designed algorithm on GPU, which allows all data required for the cross-check process to be reached without changing the reference image. The proposed method accelerates the obstacle-detection 316 times faster than the typical stereo vision. Highlights: Data fusion of ultrasonic and stereo vision is implemented. Cross-check without changing reference image is proposed. Stereo vision by parallel computing on GPU is implemented.