LWIR Target Detection Using Segmented CNN Pipelines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated target recognition (ATR) systems are not suitable for real-time use due to limitations in processing speed, video resolution, and window-based anchor box systems, especially when using convolutional neural-networks (CNNs) with long wave infrared cameras.
Innovation Solution
A high-resolution long wave infrared imaging system utilizing a custom CNN with a window-based detector for short-range targets and a pixel-based detector for longer range targets, integrated with a multi-target Kalman filter for tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a window-based detector is used for real-time detection, then processing speed is improved, but detection accuracy and target size coverage are limited
Solution Approach 1:
The patent divides the detection space into two distinct segments: a window-based detector for short-range targets and a pixel-based detector for long-range targets. This segmentation allows each detector to be optimized for its specific range, with the window-based detector providing fast processing for nearby targets and the pixel-based detector providing high accuracy for distant targets, thereby resolving the contradiction between speed and accuracy.
2Measurement precision
If a pixel-based detector is used for high accuracy detection, then detection accuracy is improved, but processing speed decreases below real-time speeds
Solution Approach 1:
The patent applies local quality by assigning different detection methods to different spatial regions: window-based detection with faster processing is applied to short-range regions, while pixel-based detection with higher accuracy is applied to long-range regions. This localized approach ensures that computational resources are allocated efficiently, providing high accuracy where needed without sacrificing real-time performance across the entire detection field.
3Measurement precision
If video resolution is increased for better target detection, then detection capability is improved, but processing complexity and computational load increase
Solution Approach 1:
The patent implements partial action by applying full-resolution pixel-based detection only to regions where long-range targets are detected, while using lower-resolution window-based detection for short-range targets. This selective approach maintains high detection capability for distant targets without requiring all processing to operate at maximum resolution, thereby reducing overall computational complexity while preserving detection effectiveness.
Data Source
AI summary
A computer program product interacts with machine-readable mediums with instructions for automated target recognition. It captures a video of a specified region using detectors on a vehicle. Raw image frames from the video undergo pre-processing, followed by feeding through long and short-range target detection pipelines. Image frames are downscaled in the short-range pipeline, generating image windows and applying a trained convolutional neural network (CNN). Full resolution and non-redundant image chips are created. Detection region of interest (ROI) proposal lists are generated and analyzed with the CNN, producing frame detection lists with detected targets. These lists are stacked and processed using a multi-target Kalman filter, ultimately creating a track list of targets for monitoring.