前言

选修的是蔡mj老师的计算机视觉，上课还是不错的，但是OpenCV可能需要自己学才能完整把作业写出来。由于没有认真学，这门课最后混了80多分，所以下面作业解题过程均为自己写的，并不是标准答案，仅供参考

任务1

修改test-3.py的task_one()函数，基于pedestrian.avi进行稀疏光流估计，对行人轨迹进行跟踪。具体输出要求如下：

（1）对视频中跟踪的轨迹进行可视化，将所有的轨迹重叠显示在视频的最后一张图像上，可视化结果保存为trajectory.png。

def task_one():"""sparse optical flow and trajectory tracking"""cap = cv2.VideoCapture("pedestrian.avi")# --------Your code--------# cap = cv2.VideoCapture("images/kk 2022-01-23 18-21-21.mp4")#cap = cv2.VideoCapture(0)# 定义角点检测的参数feature_params = dict(maxCorners=100,  # 最多多少个角点qualityLevel=0.3,  # 品质因子，在角点检测中会使用到，品质因子越大，角点质量越高，那么过滤得到的角点就越少minDistance=7  # 用于NMS，将最有可能的角点周围某个范围内的角点全部抑制)# 定义 lucas kande算法的参数lk_params = dict(winSize=(10, 10),  # 这个就是周围点临近点区域的范围maxLevel=2  # 最大的金字塔层数)# 拿到第一帧的图像ret, prev_img = cap.read()prev_img_gray = cv2.cvtColor(prev_img, cv2.COLOR_BGR2GRAY)# 先进行角点检测，得到关键点prev_points = cv2.goodFeaturesToTrack(prev_img_gray, mask=None, **feature_params)# 制作一个临时的画布，到时候可以将新的一些画的先再mask上画出来，再追加到原始图像上mask_img = np.zeros_like(prev_img)while True:ret, curr_img = cap.read()if curr_img is None:print("video is over...")breakcurr_img_gray = cv2.cvtColor(curr_img, cv2.COLOR_BGR2GRAY)# 光流追踪下curr_points, status, err = cv2.calcOpticalFlowPyrLK(prev_img_gray,curr_img_gray,prev_points,None,**lk_params)# print(status.shape)  # 取值都是1/0, 1表示是可以追踪到的，0表示失去了追踪的。good_new = curr_points[status == 1]good_old = prev_points[status == 1]# 绘制图像for i, (new, old) in enumerate(zip(good_new, good_old)):a, b = new.ravel()c, d = old.ravel()mask_img = cv2.line(mask_img, pt1=(int(a), int(b)), pt2=(int(c), int(d)), color=(0, 0, 255), thickness=1)mask_img = cv2.circle(mask_img, center=(int(a), int(b)), radius=2, color=(255, 0, 0), thickness=2)# 将画布上的图像和原始图像叠加，并且展示img = cv2.add(curr_img, mask_img)#cv2.imshow("desct", img)if cv2.waitKey(60) & 0xFF == ord('q'):print("Bye...")break# 更新下原始图像，以及重新得到新的点prev_img_gray = curr_img_gray.copy()prev_points = good_new.reshape(-1, 1, 2)if len(prev_points) < 5:# 当匹配的太少了，就重新获得当前图像的角点prev_points = cv2.goodFeaturesToTrack(curr_img_gray, mask=None, **feature_params)mask_img = np.zeros_like(prev_img)  # 重新换个画布cv2.imwrite("trajectory.png", img)

任务2

修改test-3.py的task_two()函数，基于frame01.png和frame02.png进行稠密光流估计，并基于光流估计对图像中的行人进行图像分割。具体输出要求如下：

（1）将稠密光流估计的结果进行可视化，可视化结果保存为frame01_flow.png

（2）对行人分割结果进行可视化，得到一个彩色掩码图，每个行人的分割区域用单一的颜色表示（例如red，green，blue），可视化结果保存为frame01_person.png

第二题的第一问的可视化我不清楚题目要问的是什么意思，所以跑出了两种结果。

第一种结果是背景人物分割，移动的人物会被标记为白色，背景会被标记为黑色的
第二种就是frame02图片原照片

def task_two():"""dense optical flow and pedestrian segmentation"""img1 = cv2.imread('frame01.png')img2 = cv2.imread('frame02.png')# --------Your code--------#cap = cv.VideoCapture(cv.samples.findFile("vtest.avi"))fgbg = cv2.createBackgroundSubtractorMOG2()frame1 = img1prvs = cv2.cvtColor(frame1, cv2.COLOR_BGR2GRAY)hsv = np.zeros_like(frame1)hsv[..., 1] = 255frame2 = img2fgmask = fgbg.apply(frame2)next = cv2.cvtColor(frame2, cv2.COLOR_BGR2GRAY)flow = cv2.calcOpticalFlowFarneback(prvs, next, None, 0.5, 3, 15, 3, 5, 1.2, 0)mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1])hsv[..., 0] = ang * 180 / np.pi / 2hsv[..., 2] = cv2.normalize(mag, None, 0, 255, cv2.NORM_MINMAX)bgr = cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR)#cv2.imshow('frame2', bgr)cv2.imwrite('frame01_flow.png', fgmask)cv2.imwrite('frame01_person.png', bgr)# cv2.imwrite("frame01_flow.png", img_flow)# cv2.imwrite("frame01_person.png", img_mask)

源代码

# -*- coding: utf-8 -*-
"""
Created on Mon May 29 15:30:41 2023@author: cai-mj
"""import numpy as np
import cv2
from matplotlib import pyplot as plt
import argparsedef task_one():"""sparse optical flow and trajectory tracking"""cap = cv2.VideoCapture("pedestrian.avi")# --------Your code--------# cap = cv2.VideoCapture("images/kk 2022-01-23 18-21-21.mp4")#cap = cv2.VideoCapture(0)# 定义角点检测的参数feature_params = dict(maxCorners=100,  # 最多多少个角点qualityLevel=0.3,  # 品质因子，在角点检测中会使用到，品质因子越大，角点质量越高，那么过滤得到的角点就越少minDistance=7  # 用于NMS，将最有可能的角点周围某个范围内的角点全部抑制)# 定义 lucas kande算法的参数lk_params = dict(winSize=(10, 10),  # 这个就是周围点临近点区域的范围maxLevel=2  # 最大的金字塔层数)# 拿到第一帧的图像ret, prev_img = cap.read()prev_img_gray = cv2.cvtColor(prev_img, cv2.COLOR_BGR2GRAY)# 先进行角点检测，得到关键点prev_points = cv2.goodFeaturesToTrack(prev_img_gray, mask=None, **feature_params)# 制作一个临时的画布，到时候可以将新的一些画的先再mask上画出来，再追加到原始图像上mask_img = np.zeros_like(prev_img)while True:ret, curr_img = cap.read()if curr_img is None:print("video is over...")breakcurr_img_gray = cv2.cvtColor(curr_img, cv2.COLOR_BGR2GRAY)# 光流追踪下curr_points, status, err = cv2.calcOpticalFlowPyrLK(prev_img_gray,curr_img_gray,prev_points,None,**lk_params)# print(status.shape)  # 取值都是1/0, 1表示是可以追踪到的，0表示失去了追踪的。good_new = curr_points[status == 1]good_old = prev_points[status == 1]# 绘制图像for i, (new, old) in enumerate(zip(good_new, good_old)):a, b = new.ravel()c, d = old.ravel()mask_img = cv2.line(mask_img, pt1=(int(a), int(b)), pt2=(int(c), int(d)), color=(0, 0, 255), thickness=1)mask_img = cv2.circle(mask_img, center=(int(a), int(b)), radius=2, color=(255, 0, 0), thickness=2)# 将画布上的图像和原始图像叠加，并且展示img = cv2.add(curr_img, mask_img)#cv2.imshow("desct", img)if cv2.waitKey(60) & 0xFF == ord('q'):print("Bye...")break# 更新下原始图像，以及重新得到新的点prev_img_gray = curr_img_gray.copy()prev_points = good_new.reshape(-1, 1, 2)if len(prev_points) < 5:# 当匹配的太少了，就重新获得当前图像的角点prev_points = cv2.goodFeaturesToTrack(curr_img_gray, mask=None, **feature_params)mask_img = np.zeros_like(prev_img)  # 重新换个画布cv2.imwrite("trajectory.png", img)def task_two():"""dense optical flow and pedestrian segmentation"""img1 = cv2.imread('frame01.png')img2 = cv2.imread('frame02.png')# --------Your code--------#cap = cv.VideoCapture(cv.samples.findFile("vtest.avi"))fgbg = cv2.createBackgroundSubtractorMOG2()frame1 = img1prvs = cv2.cvtColor(frame1, cv2.COLOR_BGR2GRAY)hsv = np.zeros_like(frame1)hsv[..., 1] = 255frame2 = img2fgmask = fgbg.apply(frame2)next = cv2.cvtColor(frame2, cv2.COLOR_BGR2GRAY)flow = cv2.calcOpticalFlowFarneback(prvs, next, None, 0.5, 3, 15, 3, 5, 1.2, 0)mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1])hsv[..., 0] = ang * 180 / np.pi / 2hsv[..., 2] = cv2.normalize(mag, None, 0, 255, cv2.NORM_MINMAX)bgr = cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR)#cv2.imshow('frame2', bgr)cv2.imwrite('frame01_flow.png', fgmask)cv2.imwrite('frame01_person.png', bgr)# cv2.imwrite("frame01_flow.png", img_flow)# cv2.imwrite("frame01_person.png", img_mask)if __name__ == '__main__':task_one()task_two()