In this paper we introduce a method to detect vehicle axles in an image using CNN network which is assisted by circular Hough transform and classify vehicles based on the axle count. Previous work has failed to address the issue of detecting axles for the wide field of view. To tackle this problem, we first do multi-object detection and then further process them to detect wheels. We complement a traditional Hough-based wheel detection algorithm with CNN network to reduce the computational load and the search space, further improving the efficiency. We integrate spatial and contextual filters for systematic removal of false positive wheel detection. We used MIO data set for training our vehicle detection network. We sampled 2500 images from 20 videos to create data set for our wheel detection network. We did holdout cross validation to check vehicle and wheel detection accuracy. The pipeline is simple and robust for wheel detection in wide field of view, and computationally efficient. We test the method over 5 axle classes and got average accuracy of 89.4%.
Circular Hough Transform Assisted CNN Based Vehicle Axle Detection and Classification
2019-09-01
2135995 byte
Conference paper
Electronic Resource
English
Hough Transform For Vehicle Identification
IEEE | 1993
|Hough Transform for Vehicle Identification
British Library Conference Proceedings | 1993
|A Method for Detection of Circular Arcs Based on the Hough Transform
British Library Online Contents | 1992
|Fast Hough transform-based road markings detection for autonomous vehicle
British Library Conference Proceedings | 2021
|Hough transform for long chirp detection
IEEE | 2002
|