Python 输出中的鲁棒局部光流（RLOF）坐标不正确（doc SparseRLOFOpticalFlow）

我尝试使用鲁棒局部光流（RLOF）函数（）的opencv库。为此，我使用opencv的FAST函数提取视频（帧）的不同特征点。我的视频大小为640*480，每秒30帧。然后，在整个视频中，我在第n帧和第n-1帧的主函数中使用了FAST函数。然后，我使用RLOF上的特征点检测与快速功能对图像n et图像n-1功能与不同的参数，我可以找到通过阅读该算法的文件

此外，在每一帧之间，检测到的特征点数量并不完全相同，因此我认为这就是为什么选项

setUseInitialFlow（True）

，会导致opencv

cv2错误。错误：OpenCV（4.4.0）/tmp/pip-req-build-b_zf9wbm/OpenCV_contrib/modules/optflow/src/rloffflow.cpp:372：错误：（.215：断言失败）nextPtsMat.checkVector（2，CV_32F，true）=函数“calc”中的npoints

我的代码运行正常，我的变量keypoints_old将其作为FAST的坐标（超过500个）：

但是，我用RLOF函数找到的坐标，用于找到下图中两幅图像之间的运动量，实际上并不一致。在opencv文档中，我应该从中获取点坐标。也许，我认为，这些坐标代表了运动的数量，但是，阅读文档，它不应该是这样的

keypoints_new = [[-1.0420260e+20  4.5582838e-41]
 [-1.0420260e+20  4.5582838e-41]
 [ 3.5204788e-38  0.0000000e+00]
 ...
 [-3.5464977e-24  4.5582838e-41]
 [-3.5465167e-24  4.5582838e-41]
 [-3.5465356e-24  4.5582838e-41]] 

我的问题是，如何正确使用这个函数？为什么第二幅图像的发现点坐标与前一幅图像（第一幅）的坐标不符合逻辑？我的代码怎么了？谢谢你的回答

我的职能：

import numpy as np
import cv2 as cv
import os
import argparse
import re
import os
import argparse

def FAST(img):

    # Initiate FAST object with default values

    fast = cv.FastFeatureDetector_create()
    keypoints = fast.detect(img,None)
    pts = cv.KeyPoint_convert(keypoints)
    return pts

def RLOF(old_frame, new_frame, keypoints_old, keypoints_new):

    # Parameters for RLOF

    instance = cv.optflow.RLOFOpticalFlowParameter_create()
    # instance.setUseInitialFlow(True)
    instance.setMaxIteration(30)
    instance.setNormSigma0(3.2)
    instance.setNormSigma1(7.0)
    instance.setLargeWinSize(21)
    instance.setSmallWinSize(9)
    instance.setMaxLevel(9)
    instance.setMinEigenValue(0.0001)
    instance.setCrossSegmentationThreshold(25)
    instance.setGlobalMotionRansacThreshold(10.0)
    instance.setUseIlluminationModel(True)
    instance.setUseGlobalMotionPrior(False)

    keypoints_new, st, err = cv.optflow.calcOpticalFlowSparseRLOF(
        old_frame,
        new_frame,
        keypoints_old, None, rlofParam = instance,
        forwardBackwardThreshold = 0)

    return keypoints_new

def __main__():
    parser = argparse.ArgumentParser(description='Process some video.')
    parser.add_argument('file_path', type=str,
                    help='Video_file path')
    args = parser.parse_args()
    video_file = args.file_path
    cap = cv.VideoCapture(video_file)
    fps = cap.get(cv.CAP_PROP_FPS)
    frame_count = int(cap.get(cv.CAP_PROP_FRAME_COUNT))
    ret, new_frame = cap.read() # 0
    ret, new_frame = cap.read() # 1
    p1 = FAST(new_frame)
    print('frame_count = {}'.format(frame_count))
    while cap.isOpened():
        for i in range(frame_count):
            old_frame = new_frame # 1 2 3 4
            p0 = p1
            ret, new_frame = cap.read() # 2 3 4 5 ..
            if ret:
                p1 = FAST(new_frame)
                RLOF(old_frame, new_frame, p0, p1, i)
            else:
                cap.release()
                break
__main__()

---

您有用于跟踪的图像的示例吗？

import numpy as np
import cv2 as cv
import os
import argparse
import re
import os
import argparse

def FAST(img):

    # Initiate FAST object with default values

    fast = cv.FastFeatureDetector_create()
    keypoints = fast.detect(img,None)
    pts = cv.KeyPoint_convert(keypoints)
    return pts

def RLOF(old_frame, new_frame, keypoints_old, keypoints_new):

    # Parameters for RLOF

    instance = cv.optflow.RLOFOpticalFlowParameter_create()
    # instance.setUseInitialFlow(True)
    instance.setMaxIteration(30)
    instance.setNormSigma0(3.2)
    instance.setNormSigma1(7.0)
    instance.setLargeWinSize(21)
    instance.setSmallWinSize(9)
    instance.setMaxLevel(9)
    instance.setMinEigenValue(0.0001)
    instance.setCrossSegmentationThreshold(25)
    instance.setGlobalMotionRansacThreshold(10.0)
    instance.setUseIlluminationModel(True)
    instance.setUseGlobalMotionPrior(False)

    keypoints_new, st, err = cv.optflow.calcOpticalFlowSparseRLOF(
        old_frame,
        new_frame,
        keypoints_old, None, rlofParam = instance,
        forwardBackwardThreshold = 0)

    return keypoints_new

def __main__():
    parser = argparse.ArgumentParser(description='Process some video.')
    parser.add_argument('file_path', type=str,
                    help='Video_file path')
    args = parser.parse_args()
    video_file = args.file_path
    cap = cv.VideoCapture(video_file)
    fps = cap.get(cv.CAP_PROP_FPS)
    frame_count = int(cap.get(cv.CAP_PROP_FRAME_COUNT))
    ret, new_frame = cap.read() # 0
    ret, new_frame = cap.read() # 1
    p1 = FAST(new_frame)
    print('frame_count = {}'.format(frame_count))
    while cap.isOpened():
        for i in range(frame_count):
            old_frame = new_frame # 1 2 3 4
            p0 = p1
            ret, new_frame = cap.read() # 2 3 4 5 ..
            if ret:
                p1 = FAST(new_frame)
                RLOF(old_frame, new_frame, p0, p1, i)
            else:
                cap.release()
                break
__main__()