Open AccessFront Obstacle Detection System based on YOLOv3
Author(s) -
Chao Wang,
Yan Huang,
Bowen Shi
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1732/1/012112
Subject(s) - obstacle , computer science , artificial intelligence , object detection , computer vision , set (abstract data type) , front (military) , task (project management) , field (mathematics) , position (finance) , pattern recognition (psychology) , engineering , mathematics , geography , finance , economics , pure mathematics , programming language , mechanical engineering , archaeology , systems engineering
The task of front obstacle detection in unmanned driving is to allow the model to accurately detect types and corresponding position of each object. Complex true driving environment, multiple types of objects within range of visibility and multiple examples are challenges for detection. Particularly in unmanned driving field, not only the detection precision shall be pursued, but also the detection speed shall be guaranteed. This paper puts forward the front obstacle detection method based on YOLOv3. It establishes a Tensor Flow+Keras framework on GPU server cluster, designs and realizes YOLOv3, and realizes calibration and classification of front obstacles through training and optimization of the model on the data set. This method supports over 10 object types. Experimental results show validation set Loss (VAL Loss) reaches 21.206, mAP reaches 78.64, and detection speed reaches 30 FPS. The study provides theoretical basis for obstacles detection of automated vehicle in complex environment.