复杂背景下的行人检测与分割

传统的目标检测算法在复杂环境下受到背景因素的干扰，分割出来的目标往往不能满足后期处理的需要；由于分割的好坏直接影响后期的目标跟踪的实时性和精确性的高低。鉴于此，在进行图像跟踪和识别之前，先对目标进行检测和精准的分割，提出了在AdaBoost算法中在原始Harr_like特征的基础上添加梯形特征，检测出目标的大致位置，将其作为蛇形分割的初始位置，改进蛇形分割的能量函数，分割的行人边界逐步进行收缩直至能量最小，提取出行人的真正区域。对比性实验表明改进后的算法满足实时性要求和精度要求，在一定程度上达到智能化的需求。

Pedestrian detection and segmentation under background clutter

It is difficult to get the pedestrian by the traditional algorithm of detecting objection. The segmented object won’t satisfy the later processing. The result of the segmentation affects the real-time and accuracy of the objects’tracking; For that reason, this paper detects and segments an object before tracking and recognition, adds new trapezoidal feature in Adaboost basic feature, roughly detects the position of the object and regards it as the initial location of ACM and improves the energy function. Contour of pedestrian gradually shrinks until the energy function decreases to a minimum. It extracts the real pedestrian target region from background region. Compared experiment results indicate the improved method satisfies obviously real time requirement and precision requirement. To a certain extent, it also reaches the needs of the intelligent.