ROI Adaptive Segmentation Method Based on Vanishing Point and Vehicle Height
Lifu Li, Wenying Zhang, Hui Min Zhou, Jinwen Hu · 2018 IEEE 4th Information Technology and Mechatronics Engineering Conference (ITOEC) · 2018
In image-based vehicle and pedestrian detection, the region of interest (ROI) divided by fixed ratio/region has poor adaptabilities to changeable road gradient and vehicle body pitch posture, causing excess wasted ROI or incomplete target segmentation of vehicles and pedestrians. Therefore, an ROI adaptive segmentation method based on vanishing point and vehicle height was proposed. Firstly, the method used vanishing point of lane line to get road location, avoiding excess wasted ROI. Secondly, it combined real vehicle height and detection distance to calculate vehicle height in images, complementing ROI boundaries to reduce incomplete target segmentation. Finally, the method used current frame's vanishing point and its derived real-time pitch angle to update next frame's ROI, adapting to road gradient and vehicle body pitch posture by recycling the segmentation-detection loop. Results show the method can get accurate ROI quickly under different conditions, improving real-time performance and accuracy of subsequent target detection.