在做android摄像头捕获时,发现从android摄像头出来的原始视是逆时针旋转了90度的,所以须要把它顺时针旋转90。android视频支持的是NV21格式,它是一种YUV420的格式。当然,始果你用的是android sdk的话,当中image就提供这个能力。可是我是在ndk下开发,没有找到对应的功能(假设你知道请告诉我)。
我本想用开源的图像处理库(opencv)做旋转,可是opencv仅仅能处理bmp的图像。这种话,须要先把NV21转换成BMP32。然后再做旋转。所以要操作两次,效率肯定低。最后也没找到好的方法(假设你知道一次用opencv做YUV420格式的旋转,请你告诉我)。所以仅仅有自己写一个。
先从网上搜索了一个:
void CTool::YUV420spRotate90(uchar *des, uchar *src,int width,int height) { int wh = width * height; //旋转Y int k = 0; for(int i = 0; i < width; i++) { for(int j = 0; j < height; j++) { des[k] = src[width * j + i]; k++; } } for(int i = 0; i < width; i += 2) { for(int j = 0; j < height / 2; j++) { des[k] = src[wh+ width * j + i]; des[k+1] = src[wh + width * j + i + 1]; k+=2; } } }
在android上执行了一下,发现视频很卡。所以仅仅有自己优化了。这段代码里有乘法,优化的重点就是去掉乘法用加法取代:
void CTool::YUV420spRotate90(uchar *dst, const uchar *src, int srcWidth, int srcHeight) { static int nWidth = 0, nHeight = 0; static int wh = 0; static int uvHeight = 0; if(srcWidth != nWidth || srcHeight != nHeight) { nWidth = srcWidth; nHeight = srcHeight; wh = srcWidth * srcHeight; uvHeight = srcHeight >> 1;//uvHeight = height / 2 } //旋转Y int k = 0; for(int i = 0; i < srcWidth; i++) { int nPos = 0; for(int j = 0; j < srcHeight; j++) { dst[k] = src[nPos + i]; k++; nPos += srcWidth; } } for(int i = 0; i < srcWidth; i+=2){ int nPos = wh; for(int j = 0; j < uvHeight; j++) { dst[k] = src[nPos + i]; dst[k + 1] = src[nPos + i + 1]; k += 2; nPos += srcWidth; } } return; }
逆时针旋转90度:
void CTool::YUV420spRotateNegative90(uchar *dst, const uchar *src, int srcWidth, int height) { static int nWidth = 0, nHeight = 0; static int wh = 0; static int uvHeight = 0; if(srcWidth != nWidth || height != nHeight) { nWidth = srcWidth; nHeight = height; wh = srcWidth * height; uvHeight = height >> 1;//uvHeight = height / 2 } //旋转Y int k = 0; for(int i = 0; i < srcWidth; i++){ int nPos = srcWidth - 1; for(int j = 0; j < height; j++) { dst[k] = src[nPos - i]; k++; nPos += srcWidth; } } for(int i = 0; i < srcWidth; i+=2){ int nPos = wh + srcWidth - 1; for(int j = 0; j < uvHeight; j++) { dst[k] = src[nPos - i - 1]; dst[k + 1] = src[nPos - i]; k += 2; nPos += srcWidth; } } return; }
编译执行,捕获视频显示很流畅 :)