针对车辆与行人检测的ROI自适应分割算法

在基于图像的车辆与行人检测中,针对固定比例/区域的感兴趣区域(Region of Interest, ROI)图像分割适应性低之问题,提出基于消失点和车辆高度的ROI自适应分割算法：首先,利用车道消失点得出道路位置,避免分割区域浪费；其次,综合车辆实际高度和检测距离计算图像上车辆高度,定位ROI边界,减少车辆及行人目标的不完整分割；最后,循环利用当前帧的车道消失点及其推导的实时俯仰角更新下一帧ROI,实时适应不断变化的路面坡度及车身俯仰姿态。实验表明,该算法计算简单,适应性强,满足不同情况下快速精确的ROI分割要求,提高后续目标检测的实时性和准确性。

Adaptive ROI Segmentation Algorithm for Vehicle and Pedestrian Detection

For vehicle and pedestrian detection, an adaptive region of interest (ROI) segmentation algorithm based on vanishing point (VP) and automotive height is proposed to solve the problem of low adaptabilities by constant ratio/region ROI segmentation. Initially, the VP of lane is used to get the road position and prevent waste segmentation area. The next, automotive height in images is obtained by integrating the actual automotive height and detection distance, locating the ROI border to reduce incomplete segmentation of vehicles or pedestrian. Finally, the next ROI is updated by using VP of current frame and its real-time pitch angle, which could adapt to the changing land slope and automotive body pitch in real-time. Experiment results showed the simple and effective algorithm could meet requirements of ROI segmentation in different situations, which is propitious to the real-time and accuracy of subsequent vehicle and pedestrian detection.