广东专业做网站排名公司,WordPress速度优化2019,牛街网站建设,服务器可以做自己网站用吗光流基本概念
光流表示的是相邻两帧图像中每个像素的运动速度和运动方向。具体#xff1a;光流是空间运动物体在观察成像平面上的像素运动的瞬时速度#xff0c;是利用图像序列中像素在时间域上的变化以及相邻帧之间的相关性来找到上一帧跟当前帧之间存在的对应关系#xf…光流基本概念
光流表示的是相邻两帧图像中每个像素的运动速度和运动方向。具体光流是空间运动物体在观察成像平面上的像素运动的瞬时速度是利用图像序列中像素在时间域上的变化以及相邻帧之间的相关性来找到上一帧跟当前帧之间存在的对应关系从而计算出相邻帧之间物体的运动信息的一种方法。一般而言光流是由于场景中前景目标本身的移动、相机的运动或者两者的共同运动所产生的。
opencv光流算子
参考链接
光流法大全(DeepFlow、DenseFlow、DisFlow、FbFlow、PCAFlow、SimpleFlow、TV_L1)
视频光流计算demo
import os
import cv2
import flow_vis
import numpy as np
from tqdm import tqdmdef compute_flow(prev, curr, bound15):Farneback optical flow# flow cv2.calcOpticalFlowFarneback(prevprev, nextcurr, flowNone, pyr_scale0.5, levels5,winsize15, iterations3, poly_n5, poly_sigma1.1, flagscv2.OPTFLOW_FARNEBACK_GAUSSIAN) TVL1 optical flow需安装opencv_contribTVL1 cv2.optflow.DualTVL1OpticalFlow_create()flow TVL1.calc(prev, curr, None) deepflow optical flow 需安装opencv_contrib# inst cv2.optflow.createOptFlow_DeepFlow()# flow inst.calc(prev, curr, None) sparse to dense flow optical flow 需安装opencv_contrib# flow cv2.optflow.calcOpticalFlowSparseToDense(prev, curr) pca flow optical flow 需安装opencv_contrib# inst cv2.optflow.createOptFlow_PCAFlow()# flow inst.calc(prev, curr, None) DIS optical flow # dis cv2.DISOpticalFlow_create(2)# flow dis.calc(prev, curr, None)return flowdef video2flow(video_path:str, flow_path:str): 读取视频获取视频基本信息 videoCapture cv2.VideoCapture(video_path)if not videoCapture.isOpened(): # 若视频文件读取失败读取下一段视频print(视频打开失败!!!)print(video_path)return Falsetotal_frames int(videoCapture.get(cv2.CAP_PROP_FRAME_COUNT)) # 获取视频总帧数# fourcc int(videoCapture.get(cv2.CAP_PROP_FOURCC)) # 原生不支持h264编码fps videoCapture.get(cv2.CAP_PROP_FPS) # 获取视频帧率w int(videoCapture.get(cv2.CAP_PROP_FRAME_WIDTH)) # 获取图像宽度h int(videoCapture.get(cv2.CAP_PROP_FRAME_HEIGHT)) # 获取图像高度光流视频 写入设置 # fourcc cv2.VideoWriter_fourcc(*MJPG) # avi格式fourcc cv2.VideoWriter_fourcc(*mp4v) # MP4格式video_flow cv2.VideoWriter(flow_path, fourcc, fps, frameSize(w, h), isColorTrue) 计算光流并保存 success, prev videoCapture.read()total_frames - 1if not success:print(视频首帧读取失败)return Falsepbar tqdm(totaltotal_frames)while total_frames:success, curr videoCapture.read()if success:prev_gray cv2.cvtColor(prev, cv2.COLOR_BGR2GRAY)curr_gray cv2.cvtColor(curr, cv2.COLOR_BGR2GRAY)tmp_flow compute_flow(prev_gray, curr_gray)rgb flow_vis.flow_to_color(tmp_flow, convert_to_bgrFalse)video_flow.write(rgb.astype(np.uint8))prev curr# cv2.imshow(frame, flow_xy.astype(np.uint8))# cv2.waitKey(0)qelse:print(某中间帧读取失败光流视频生成失败)return Falsepbar.update(1)total_frames - 1videoCapture.release()video_flow.release()cv2.destroyAllWindows()return Trueif __name____main__:video_path ./forest.mp4flow_path ./forest_flow.mp4print(video2flow(video_path, flow_path))
其中demo中flow_vis使用的光流调色板(Color wheel)如下颜色代表光流方向颜色深度代表光流速度
