ios - 保存高质量图像，进行实时处理——什么是最好的方法？

Question

我仍在学习 AVFoundation，所以我不确定我应该如何最好地解决需要捕获高质量静止图像但提供低质量预览视频流的问题。

我有一个应用程序需要拍摄高质量图像 (AVCaptureSessionPresetPhoto)，但使用 OpenCV 处理预览视频流 - 对于低得多的分辨率是可以接受的。简单地使用基础 OpenCV Video Camera class不好，因为将 defaultAVCaptureSessionPreset 设置为 AVCaptureSessionPresetPhoto 会导致将全分辨率帧传递给 processImage - 这确实非常慢。

如何与可用于捕获静止图像的设备建立高质量连接，以及如何处理和显示低质量连接？关于我需要如何设置 session /连接的描述将非常有帮助。是否有此类应用的开源示例？

Answer 1

我做了类似的事情——我在委托(delegate)方法中抓取像素，制作它们的 CGImageRef，然后将其分派(dispatch)到正常的优先级队列，在那里它被修改。由于 AVFoundation 必须使用 CADisplayLink 作为回调方法，因此它具有最高优先级。在我的特殊情况下，我没有抓取所有像素，因此它可以在 iPhone 4 上以 30fps 的速度运行。根据您要运行的设备，您需要权衡像素数、fps 等。

另一个想法是获取像素子集的 2 次幂 - 例如每行中的每 4 个和每 4 行。我再次以 20-30fps 的速度在我的应用程序中做了类似的事情。然后，您可以在分派(dispatch)的 block 中进一步操作这个较小的图像。

如果这看起来令人生畏，请悬赏工作代码。

代码:

// Image is oriented with bottle neck to the left and the bottle bottom on the right
- (void)captureOutput:(AVCaptureVideoDataOutput *)captureOutput didOutputSampleBuffer:(CMSampleBufferRef)sampleBuffer fromConnection:(AVCaptureConnection *)connection
{
#if 1   
    AVCaptureDevice *camera = [(AVCaptureDeviceInput *)[captureSession.inputs lastObject] device];
    if(camera.adjustingWhiteBalance || camera.adjustingExposure) NSLog(@"GOTCHA: %d %d", camera.adjustingWhiteBalance, camera.adjustingExposure);
    printf("foo\n");
#endif

    if(saveState != saveOne && saveState != saveAll) return;


    @autoreleasepool {
        CVImageBufferRef imageBuffer = CMSampleBufferGetImageBuffer(sampleBuffer); 
        //NSLog(@"PE: value=%lld timeScale=%d flags=%x", prStamp.value, prStamp.timescale, prStamp.flags);

        /*Lock the image buffer*/
        CVPixelBufferLockBaseAddress(imageBuffer,0); 

        NSRange captureRange;
        if(saveState == saveOne) {
#if 0 // B G R A MODE ! 
NSLog(@"PIXEL_TYPE: 0x%lx", CVPixelBufferGetPixelFormatType(imageBuffer));
uint8_t *newPtr = (uint8_t *)CVPixelBufferGetBaseAddress(imageBuffer);
NSLog(@"ONE VAL %x %x %x %x", newPtr[0], newPtr[1], newPtr[2], newPtr[3]);
}
exit(0);
#endif
            [edgeFinder setupImageBuffer:imageBuffer];

            BOOL success = [edgeFinder delineate:1];

            if(!success) {
                dispatch_async(dispatch_get_main_queue(), ^{ edgeFinder = nil; [delegate error]; });
                saveState = saveNone;
            } else 
                bottleRange = edgeFinder.sides;
                xRange.location = edgeFinder.shoulder;
                xRange.length = edgeFinder.bottom - xRange.location;

                NSLog(@"bottleRange 1: %@ neck=%d bottom=%d", NSStringFromRange(bottleRange), edgeFinder.shoulder, edgeFinder.bottom );
                //searchRows = [edgeFinder expandRange:bottleRange];

                rowsPerSwath = lrintf((bottleRange.length*NUM_DEGREES_TO_GRAB)*(float)M_PI/360.0f);
NSLog(@"rowsPerSwath = %d", rowsPerSwath);
                saveState = saveIdling;

                captureRange = NSMakeRange(0, [WLIPBase numRows]);
                dispatch_async(dispatch_get_main_queue(), ^
                    {
                        [delegate focusDone];
                        edgeFinder = nil;
                        captureOutput.alwaysDiscardsLateVideoFrames = YES;
                    });
        } else {        
            NSInteger rows = rowsPerSwath;
            NSInteger newOffset = bottleRange.length - rows;
            if(newOffset & 1) {
                --newOffset;
                ++rows;
            }
            captureRange = NSMakeRange(bottleRange.location + newOffset/2, rows);
        }
        //NSLog(@"captureRange=%u %u", captureRange.location, captureRange.length);

        /*Get information about the image*/
        uint8_t *baseAddress = (uint8_t *)CVPixelBufferGetBaseAddress(imageBuffer); 
        size_t bytesPerRow = CVPixelBufferGetBytesPerRow(imageBuffer); 
        size_t width = CVPixelBufferGetWidth(imageBuffer); 

        // Note Apple sample code cheats big time - the phone is big endian so this reverses the "apparent" order of bytes
        CGContextRef newContext = CGBitmapContextCreate(NULL, width, captureRange.length, 8, bytesPerRow, colorSpace, kCGImageAlphaNoneSkipFirst | kCGBitmapByteOrder32Little); // Video in ARGB format

assert(newContext);

        uint8_t *newPtr = (uint8_t *)CGBitmapContextGetData(newContext);
        size_t offset   = captureRange.location * bytesPerRow;

        memcpy(newPtr, baseAddress + offset, captureRange.length * bytesPerRow);

        CVPixelBufferUnlockBaseAddress(imageBuffer, 0);

        OSAtomicIncrement32(&totalImages);
        int32_t curDepth = OSAtomicIncrement32(&queueDepth);
        if(curDepth > maxDepth) maxDepth = curDepth;

#define kImageContext   @"kImageContext"
#define kState          @"kState"
#define kPresTime       @"kPresTime"

        CMTime prStamp = CMSampleBufferGetPresentationTimeStamp(sampleBuffer);      // when it was taken?
        //CMTime deStamp = CMSampleBufferGetDecodeTimeStamp(sampleBuffer);          // now?

        NSDictionary *dict = [NSDictionary dictionaryWithObjectsAndKeys:
            [NSValue valueWithBytes:&saveState objCType:@encode(saveImages)], kState,
            [NSValue valueWithNonretainedObject:(__bridge id)newContext], kImageContext,
            [NSValue valueWithBytes:&prStamp objCType:@encode(CMTime)], kPresTime,
            nil ];
        dispatch_async(imageQueue, ^
            {
                // could be on any thread now
                OSAtomicDecrement32(&queueDepth);

                if(!isCancelled) {
                    saveImages state; [(NSValue *)[dict objectForKey:kState] getValue:&state];
                    CGContextRef context; [(NSValue *)[dict objectForKey:kImageContext] getValue:&context];
                    CMTime stamp; [(NSValue *)[dict objectForKey:kPresTime] getValue:&stamp];

                    CGImageRef newImageRef = CGBitmapContextCreateImage(context); 
                    CGContextRelease(context);
                    UIImageOrientation orient = state == saveOne ? UIImageOrientationLeft : UIImageOrientationUp;
                    UIImage *image = [UIImage imageWithCGImage:newImageRef scale:1.0 orientation:orient]; // imageWithCGImage:  UIImageOrientationUp  UIImageOrientationLeft
                    CGImageRelease(newImageRef);
                    NSData *data = UIImagePNGRepresentation(image);

                    // NSLog(@"STATE:[%d]: value=%lld timeScale=%d flags=%x", state, stamp.value, stamp.timescale, stamp.flags);

                    {
                        NSString *name = [NSString stringWithFormat:@"%d.png", num];
                        NSString *path = [[wlAppDelegate snippetsDirectory] stringByAppendingPathComponent:name];
                        BOOL ret = [data writeToFile:path atomically:NO];
//NSLog(@"WROTE %d err=%d w/time %f path:%@", num, ret, (double)stamp.value/(double)stamp.timescale, path);
                        if(!ret) {
                            ++errors;
                        } else {
                            dispatch_async(dispatch_get_main_queue(), ^
                                {
                                    if(num) [delegate progress:(CGFloat)num/(CGFloat)(MORE_THAN_ONE_REV * SNAPS_PER_SEC) file:path];
                                } );
                        }
                        ++num;
                    }
                } else NSLog(@"CANCELLED");

            } );
    }
}