基礎部落格

請參考我的基於opencv模板匹配的單物體跟蹤，基本原理相同，增加了矩陣記錄多目標的位置資訊等。

import cv2 as cv2import numpy as npfrom subprocess import calldef template_demo(tpl, target, method = cv2.TM_CCORR_NORMED):    th, tw = tpl.shape[:2]# 取高寬，不取通道 模板高寬    result = cv2.matchTemplate(target, tpl, method)    min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(result) # 尋找最小值，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, lostif __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')  # 儲存影片的編碼    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())1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980