使用AVCaptureMovieFileOutput录制视频时切换摄像头的方案咨询

Great question—this is a super common pain point when building Instagram-style recording features with AVFoundation! Let’s break down your two options, plus the pros/cons of each:

1. 分段录制 + 后期合并：可行，且实现相对简单 #

This approach absolutely works, and it’s the easier of the two options if you’re looking to get a functional version up quickly. Here’s how it would work:

• When the user taps to switch cameras, first stop the current recording and wait for the AVCaptureFileOutputRecordingDelegate callback to confirm the file saved successfully (make sure to handle any errors here, like failed writes).
• Switch your AVCaptureSession’s video input: remove the current camera’s input, add the new front/back camera input, and restart the session configuration.
• Start recording a new video file with the new camera.
• Once the user finishes recording, use AVComposition and AVAssetExportSession to stitch the two (or more) video files into a single seamless clip.

Pros: #

• Uses AVCaptureMovieFileOutput’s built-in recording logic—you don’t have to handle low-level video frame processing, which reduces bugs.
• Lower learning curve if you’re already familiar with basic AVCaptureSession setup.

Cons: #

• There will be a noticeable pause between stopping one recording and starting the next—you’ll need to add UI transitions (like a crossfade or camera flip animation) to mask this, but it won’t be as smooth as Instagram’s seamless switch.
• Merging files takes extra processing time, especially for longer videos, which could delay the final video being ready for sharing.
• You need to ensure all recorded clips have matching resolution, frame rate, and bitrate to avoid glitches during merging.

2. 使用AVCaptureVideoDataOutput：实现无缝切换，但复杂度更高 #

If you want that Instagram-style seamless camera switch without stopping recording, this is the way to go. Instead of relying on AVCaptureMovieFileOutput to handle writing files, you’ll capture raw video frames directly and write them to a file using AVAssetWriter.

Here’s the core workflow:

• Pre-configure both front and back camera AVCaptureDeviceInputs and keep them ready (don’t add both to the session at once—just hold references to them).
• Use AVCaptureVideoDataOutput to capture video frames, and AVCaptureAudioDataOutput for audio (since you still need to record sound).
• When the user taps to switch cameras, perform the input swap on the capture session’s serial queue (critical to avoid UI freezes): remove the current camera input, add the new one, and the session will reconfigure without stopping the frame capture.
• Feed all incoming frames (from both cameras) into AVAssetWriter to build a single continuous video file.

Pros: #

• Completely seamless camera switching—no pause in recording, just like Instagram Stories.
• Full control over frame processing (you can add filters, transitions, or other effects mid-recording if needed).

Cons: #

• Way more complex: you’ll have to handle frame timestamp synchronization, audio-video alignment, and raw data conversion (since AVCaptureVideoDataOutput gives you unprocessed pixel buffers).
• Higher risk of bugs: missed frames, audio desync, or corrupted output files if you don’t handle the AVAssetWriter lifecycle correctly.
• You’ll need to manage the capture session’s queue carefully to avoid blocking the main thread.

Which should you choose? #

• If you’re prioritizing speed of development and simplicity, start with the segmented recording + merge approach. It’s a solid MVP solution that works well enough for most use cases.
• If seamless user experience is non-negotiable (like matching Instagram’s polish), invest the time to implement the AVCaptureVideoDataOutput + AVAssetWriter approach. It’s harder, but the end result is worth it.

Quick note on why AVCaptureMovieFileOutput stops recording when switching cameras: #

The reason this happens is that modifying the AVCaptureSession’s inputs (adding/removing a camera) triggers a session reconfiguration. AVCaptureMovieFileOutput is designed to stop recording automatically during reconfigurations—this is a system-level behavior you can’t override, which is why the segmented approach is necessary if you stick with this output type.

内容的提问来源于stack exchange，提问作者croigsalvador