视觉算法:非极大值抑制nms

0.绪:

非极大值抑制-Non Maximum Suppression,简称NMS;

经常应用在目标检测的后期=物体边界框去除;NMS对检测得到的全部 boxes 进行局部的最大搜索，以搜索某邻域范围内的最大值，从而滤处一部分 boxes，提升最终的检测精度.

1.过程:

去除冗余的重叠 Boxes，对全部的 Boxes 进行迭代-遍历-消除.

s1.将所有框的得分排序,选中最高分及其对应的框;

s2.遍历其余的框,如果和当前最高分框的交并比IOU大于设定阈值,记录此框索引;

遍历完成后，保存最高得分框，并在原集合内删除最高得分框和大于阈值的框；

s3.从未处理的框中继续选择一个得分最高的,重复上述过程.

2.输入输出:

输入:检测到的Boxes(同一个物体可能被检测到很多Boxes，每个box均有分类score);

输出:最优的Box;

3.python代码:

依赖:python2.7和opencv-python

main.py:

# -*- coding: utf-8 -*-
from __future__ import division, print_function, absolute_import
import numpy as np
import cv2
import nms


if __name__ == "__main__":
    boxes = [[50.,50.,100.,100.,0.95],
             [55.,55.,100.,100.,0.75],
             [45.,45.,100.,100.,0.83],
             [200.,50.,100.,100.,0.81],
             [195.,55.,100.,100.,0.67]]

    boxes = np.array(boxes)

    I = np.zeros((400, 400), dtype=np.uint8)
    I = cv2.cvtColor(I, cv2.COLOR_GRAY2BGR)
    I1 = I.copy()
    scores = boxes[:,4]
    for box in boxes:
        # 画矩形框
        cv2.rectangle(I, (int(box[0]), int(box[1])), (int(box[0]+box[2]), int(box[1]+box[3])), (0, 255, 0), 1)


    cv2.namedWindow("image")
    cv2.imshow("image", I)




    nms_max_overlap = 0.5
    indices = nms.non_max_suppression( boxes, nms_max_overlap, scores)
    print(indices)


    for ii in indices:
        # 画矩形框
        cv2.rectangle(I1, (int(boxes[ii,0]), int(boxes[ii,1])), (int(boxes[ii,0]+boxes[ii,2]), int(boxes[ii,1]+boxes[ii,3])), (0, 255, 0), 1)

    cv2.namedWindow("dst")
    cv2.imshow("dst", I1)
    cv2.waitKey()

nms.py:

# -*- coding: utf-8 -*-
# vim: expandtab:ts=4:sw=4
import numpy as np
import cv2


def non_max_suppression(boxes, max_bbox_overlap, scores=None):
    """Suppress overlapping detections.

    Original code from [1]_ has been adapted to include confidence score.

    .. [1] http://www.pyimagesearch.com/2015/02/16/
           faster-non-maximum-suppression-python/

    Examples
    --------

        >>> boxes = [d.roi for d in detections]
        >>> scores = [d.confidence for d in detections]
        >>> indices = non_max_suppression(boxes, max_bbox_overlap, scores)
        >>> detections = [detections[i] for i in indices]

    Parameters
    ----------
    boxes : ndarray
        Array of ROIs (x, y, width, height).
    max_bbox_overlap : float
        ROIs that overlap more than this values are suppressed.
    scores : Optional[array_like]
        Detector confidence score.

    Returns
    -------
    List[int]
        Returns indices of detections that have survived non-maxima suppression.

    """
    if len(boxes) == 0:
        return []

    boxes = boxes.astype(np.float) #将数组类型转为为float
    pick = []

    x1 = boxes[:, 0]
    y1 = boxes[:, 1]
    x2 = boxes[:, 2] + boxes[:, 0] #x1+w
    y2 = boxes[:, 3] + boxes[:, 1] #y1+h

    area = (x2 - x1 + 1) * (y2 - y1 + 1)
    if scores is not None:
        idxs = np.argsort(scores) #从小到大排序=顺序
    else:
        idxs = np.argsort(y2)

    while len(idxs) > 0:
        last = len(idxs) - 1
        i = idxs[last]
        pick.append(i)

        xx1 = np.maximum(x1[i], x1[idxs[:last]])
        yy1 = np.maximum(y1[i], y1[idxs[:last]])
        xx2 = np.minimum(x2[i], x2[idxs[:last]])
        yy2 = np.minimum(y2[i], y2[idxs[:last]])

        w = np.maximum(0, xx2 - xx1 + 1)
        h = np.maximum(0, yy2 - yy1 + 1)

        overlap = (w * h) / area[idxs[:last]]

        idxs = np.delete(
            idxs, np.concatenate(
                ([last], np.where(overlap > max_bbox_overlap)[0])))

    return pick

结果:

 

参考:

1.https://blog.youkuaiyun.com/qq_38906523/article/details/80195119

2.https://blog.youkuaiyun.com/zziahgf/article/details/78525279

3.https://blog.youkuaiyun.com/shuzfan/article/details/52711706

4.https://www.pyimagesearch.com/2015/02/16/faster-non-maximum-suppression-python/    源码所在