Combination of SSD and a Rule-Based Approach for Nighttime Vehicle Detection on Roads
Jeongmin Park, Joon-Woong Lee · Journal of Institute of Control Robotics and Systems · 2020
We propose a new approach to detect preceding and on-coming vehicles in front of an ego-vehicle from images captured on roads at night. The new approach comprises two main steps: vehicle candidate generation and candidate verification. Vehicle candidates are generated by using rule-based logics combined with SSD. This combination improves a previous approach [2] in terms of vehicle detection performance. This previous approach presents difficulties in detecting vehicles that are near but performs relatively better performance when detecting vehicles that are far. In contrast, compared with the previous approach, SSD shows a better performance in detecting vehicles that are near. However, SSD performs poorly in detecting vehicles that are far. The verification is realized by a classifier designed by VGG. The experimental results for actual road images illustrate the effectiveness of the proposed algorithm.