基于opencv的多目标物体跟踪!走到哪跟到哪！

 

基础博客

请参考我的基于opencv模板匹配的单物体跟踪，基本原理相同，增加了矩阵记录多目标的位置信息等。

代码

import cv2 as cv2
import numpy as np
from subprocess import call


def template_demo(tpl, target, method = cv2.TM_CCORR_NORMED):
    th, tw = tpl.shape[:2]# 取高宽，不取通道 模板高宽
    result = cv2.matchTemplate(target, tpl, method)
    min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(result) # 寻找最小值，最大值。最小值位置，最大值位置
    tl = max_loc
    br = (tl[0]+tw, tl[1]+th)
    print(max_val, tl, br)
    if max_val < 0.45:
        lost = 1
    else:
        lost = 0
    return tl, br, lost



if __name__ == '__main__':

    number = int(input('please input the number of objects: '))
    area = np.zeros((number, 4))
    ROI_area = np.zeros((number, 4))
    print(area.shape)
    print("--------- Python OpenCV Tutorial ---------")
    cap = cv2.VideoCapture('../robot.mp4')
    ret, frame = cap.read()
    original_frame = frame
    print(frame.shape)
    lam = 1
    fourcc = cv2.VideoWriter_fourcc(*'XVID')  # 保存视频的编码
    out = cv2.VideoWriter('football_output.avi',fourcc, 20.0, (frame.shape[1], frame.shape[0]))
    for i in range(number):
        gROI = cv2.selectROI("ROI frame", frame, False)
        ROI_area[i] = [gROI[1], gROI[1]+gROI[3], gROI[0], gROI[0]+gROI[2]]
        # ROI = frame[gROI[1]: gROI[1]+gROI[3], gROI[0]:gROI[0]+gROI[2], :]
        area[i] = [0, frame.shape[1], 0, frame.shape[0]]
    print(area)
    print(ROI_area)
    while True:
        ret, frame = cap.read()
        if ret:
            for k in range(number):
                ROI = original_frame[int(ROI_area[k, 0]): int(ROI_area[k, 1]), int(ROI_area[k, 2]):int(ROI_area[k,3]), :]
                frame1 = frame[int(area[k, 2]):int(area[k, 3]), int(area[k, 0]):int(area[k, 1]), :]
                tl, br, lost = template_demo(ROI, frame1, method=cv2.TM_CCOEFF_NORMED)
                if lost==1:
                    lam = 2
                else:
                    lam = 1
                # ROI = frame1[tl[1]:br[1], tl[0]:br[0], :]
                cv2.imshow('ROI', ROI)
                print(ROI.shape)
                result = cv2.rectangle(frame, (int(area[k, 0])+tl[0], int(area[k, 2])+tl[1]), (br[0]+int(area[k, 0]), br[1]+int(area[k, 2])), (k*20, k*10, 255//(k+1)), 2)
                

                area[k, 0] = (tl[0] + br[0]) // 2 - lam * (br[0] - tl[0]) + area[k, 0]
                area[k, 1] = (tl[0] + br[0]) // 2 + lam * (br[0] - tl[0]) + area[k, 0]
                area[k, 2] = (tl[1] + br[1]) // 2 - lam * (br[1] - tl[1]) + area[k, 2]
                area[k, 3] = (tl[1] + br[1]) // 2 + lam * (br[1] - tl[1]) + area[k, 2]
                area[k, 0] = 0 if area[k, 0] < 0 else area[k, 0]
                area[k, 1] = frame.shape[1] if area[k, 1] > frame.shape[1] else area[k, 1]
                area[k, 2] = 0 if area[k, 2] < 0 else area[k, 2]
                area[k, 3] = frame.shape[0] if area[k, 3] > frame.shape[0] else area[k, 3]

            out.write(result)
            cv2.waitKey(18)
            cv2.imshow('result', result)
        else:
            break
        
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
        
    cap.release()
    cv2.destroyAllWindows()
    command = "ffmpeg -i football_output.avi football_output.mp4"
    call(command.split())

展示

 

后记

近期有很多朋友通过私信咨询有关Python学习问题。为便于交流，点击蓝色自己加入讨论解答资源基地