Multi-Code Scanning Framework and Optimization for Mobile Apps

This article documents the redesign of a scanning feature from single‑code to multi‑code recognition, describing the technical framework, algorithmic improvements, and performance optimizations.

Background & Challenge – Multiple barcodes or QR codes often appear together (e.g., clothing tags, beverage bottles). The original scanner only handled a single code, leaving users unaware of which code was detected. The goal was to enable multi‑code detection, anchor each code’s position, and retain compatibility with single‑code scenarios.

Logic Layer – The processing pipeline was changed from handling a single data object to iterating over an array of codes. The decoder now caches the current frame for later display.

View Layer – After detection, the UI overlays arrows anchored to each code’s center on the cached frame. Two display strategies were evaluated:

Stop the camera session and keep the last frame static.

Cache the current frame as an image overlay.

Both have trade‑offs; the final solution controls memory via resolution scaling and releases cached frames after user interaction.

Code‑Box Transformation Algorithm

码框旋转rect = CGRect(x: 原图宽 - (原码框.y + 原码框.height), y: 原码框.x, width: 原码框.height, height: 原码框.width);
BOOL 是否长屏幕 = 图片宽/图片高 > 屏幕宽/屏幕高 ? YES : NO;
缩放比例 = 是否长屏幕 ? 屏幕高/图片高 : 屏幕宽/相机宽;
码框缩放rect = CGRect(x: 图片旋转rect.x * scale, y: 图片旋转rect.y * scale, width: 图片旋转rect.width * scale, height: 图片旋转rect.height * scale);
if (长屏幕) { CGFloat xOffset = (图片宽*scale - 屏幕宽)/2.0; 码框坐标变换rect = CGRectMake(x:码框缩放rect.x-xOffset, y:码框缩放rect.y, width:码框缩放rect.width, height:码框缩放rect.height); } else { CGFloat yOffset = (图片高*scale - 屏幕高)/2.0; 码框坐标变换rect = CGRectMake(x:码框缩放rect.x, y:码框缩放rect.y-yOffset, width:码框缩放rect.width, height:码框缩放rect.height); }
return 码框坐标变换rect;

Reducing Miss‑Detection Rate – Single‑frame decoding success depends on blur, compression, and algorithmic limits. Multi‑code scenarios showed higher miss rates, prompting additional strategies.

Vision Decoding – iOS 13+ Vision API was integrated in parallel with the existing SDK. Both decoders take ~50‑100 ms, allowing serial processing without blocking the UI. Vision provides supplemental detections, especially for codes missed by the SDK.

Confidence Filtering & Deduplication – Results with confidence < 0.9 are discarded. Duplicate detections (same payload and overlapping position) are merged, keeping the entry with the highest confidence.

Model Mapping & Data Fusion – Vision outputs are mapped to the SDK’s data model (type, payload, confidence, bounding box) and then fused with SDK results using the same deduplication logic.

Edge Intelligence – A lightweight on‑device code‑detection model predicts code locations before decoding, enabling targeted cropping and improving SDK success rates.

Results – The combined approach raised multi‑code recognition rates significantly (over 30 percentage points in album mode) and reduced miss‑detections across various scenarios.

Conclusion & Outlook – Continuous improvements in SDK algorithms, iOS Vision, and edge‑intelligence are essential for sustaining high‑quality scanning experiences.