Multiband Multidirectional Filtering for Slow-Moving Target Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target detection methods struggle with low accuracy in detecting moving targets that move slowly or rotate, often resulting in incomplete detection and holes in the detection process.
Innovation Solution
A method involving multiband and multidirectional filtering of video frames to obtain band and direction sub-images, followed by fusing these features for input into a detection model to enhance target detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional target detection methods are used, then the detection process is simple, but the detection accuracy is low for moving targets that move slowly or rotate
Solution Approach 1:
The patent applies segmentation by dividing the image processing into multiple frequency bands using multiband filtering. The image is decomposed into different frequency components (low-frequency, medium-frequency, high-frequency bands), allowing the detection system to analyze target features at different scales and orientations separately, thereby improving detection accuracy for slowly moving or rotating targets
Solution Approach 2:
The patent introduces directional filtering to add an orientation dimension to the detection process. By filtering images in multiple directions (horizontal, vertical, diagonal orientations) and fusing the results, the system captures target features from multiple angular perspectives, significantly improving the ability to detect rotating targets
2Measurement precision
If multiband and multidirectional filtering is applied, then detection accuracy improves, but computational complexity increases
Solution Approach 1:
The patent performs preliminary multiband filtering to decompose the image into frequency components before directional filtering. This preliminary segmentation allows subsequent directional filtering to operate on already-simplified frequency bands, reducing the overall computational burden compared to applying directional filtering on the full-resolution image
Solution Approach 2:
The patent merges the results from multiple directional filters through feature fusion. Instead of processing each direction separately through complete detection pipelines, the system combines directional features at the feature level, reducing redundant computations while maintaining detection accuracy
Data Source
AI summary
A detection method includes performing multiband filtering on a first area to obtain a plurality of band sub-images, the first area being an area in a first video frame, and performing multidirectional filtering on the plurality of band sub-images to obtain a plurality of direction sub-images. The method further includes acquiring a direction-band fused feature of the first area according to the plurality of direction sub-images, and inputting the direction-band fused feature into a detection model, and performing detection based on the direction-band fused feature using the detection model to detect whether the first area comprises an object.