Multi-Scale Video Target Acquisition via Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-target tracking techniques are limited by low classification precision and complex algorithms, making them inefficient for real-time video applications, and struggle to accurately distinguish between multiple targets and their backgrounds.
Innovation Solution
A target acquisition method that involves obtaining pre-estimated positions of multiple scales in video frames using global features, clustering these positions to form candidate regions, and determining actual regions based on confidence levels, with features including grayscale, texture, color, and motion analysis, and multi-scale frequency domain filtering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for multi-class classification, then classification precision is improved, but algorithm complexity increases and operation efficiency decreases
Solution Approach 1:
The patent segments the target detection task into multiple independent single-target detection tasks at different scales. Instead of using a complex multi-class classifier to distinguish between multiple target types simultaneously, the system performs separate detection for each scale level (e.g., small, medium, large targets), then combines the results. This segmentation approach simplifies each individual detection task while maintaining overall accuracy.
Solution Approach 2:
The patent introduces a multi-scale dimension to the target detection problem. Rather than attempting to classify targets into multiple categories using complex neural networks, the system processes targets at different spatial scales independently. This dimensional transformation from temporal/spectral classification to spatial scale separation simplifies the algorithm while improving efficiency.
2Reliability
If conventional multi-target tracking algorithms are applied, then tracking capability is achieved, but computational complexity increases and real-time processing becomes difficult
Solution Approach 1:
The patent segments the video processing into independent frames and processes each frame separately at different scales. By dividing the complex multi-target tracking problem into independent single-frame detection tasks, the system avoids the computational burden of tracking algorithms that require processing multiple frames simultaneously, enabling real-time operation.
Solution Approach 2:
The patent performs preliminary detection at multiple scales for each frame before any tracking occurs. By pre-identifying potential targets at different scales in each individual frame, the system prepares the data needed for subsequent tracking without requiring complex real-time multi-target discrimination algorithms.
3Adaptability or versatility
If DNN-based multi-class models are trained offline, then comprehensive target detection is achieved, but training time and computational resources increase significantly
Solution Approach 1:
Instead of training a comprehensive multi-class DNN model that requires significant computational resources and time, the patent uses simple, lightweight single-target detectors at each scale. These simple detectors can be trained quickly and with minimal computational resources, yet collectively provide comprehensive target detection capability across different scales.
Solution Approach 2:
The patent segments the comprehensive target detection task into multiple simple single-scale detection tasks. Each scale level uses a simple detector that can be trained independently and quickly, avoiding the need for a single complex model that would require extensive training time and computational resources.
Data Source
AI summary
An example target acquisition method includes obtaining, according to a global feature of each video frame of a plurality of video frames, a target pre-estimated position of each scale in the video frame; clustering the target pre-estimated position in each video frame to obtain a corresponding target candidate region; and determining a target actual region in the video frame according to all the target candidate regions in each video frame in combination with confidence levels of the target candidate regions and corresponding scale processing. The techniques of the present disclosure quickly and effectively acquire one or multiple targets, and, more particularly, achieve accurately distinguishing and acquiring the multiple targets.