1. Vehicle Positioning Method of Roadside Monocular Camera
- Author
-
Xinhong Wang, Ping Wang, and Ren Xiaoyang
- Subjects
Computer science ,business.industry ,Position (vector) ,Minimum bounding box ,Frame (networking) ,Wireless ,Computer vision ,Artificial intelligence ,business ,Parallelogram ,Convolutional neural network ,Wireless sensor network ,Camera resectioning - Abstract
In this work, we propose an end-to-end detection method based on the 2D bounding box detection to locate vehicles with a roadside monocular camera. Our work is based on the fact that the shape of the vehicles bottom plane frame in the RGB image of roadside monocular camera is approximately parallelogram, which is validated by the dataset BoxCars116k. Compared with previous work in literature, the method takes full advantage of key points detection by convolutional neural network, which can estimate the vehicle position more simpler and efficiently. Furthermore, the method can make more accurate location for the vehicles on the road by camera calibration. The performance of our end-to-end detection method is validated by the experimental results on the BoxCars116k dataset.
- Published
- 2020