Two-Stage Object Detection for Fast-Moving Target Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection and tracking methods are computationally expensive and struggle with small or distant objects, especially when they move quickly and erratically, requiring significant computational resources and often failing to provide accurate results.
Innovation Solution
A method involving a two-step detection process, where a first detection step analyzes the entire image for rough object localization and a second, more precise step focuses on selected sub-images for verification, combined with sub-image tracking and confidence-based failure recovery mechanisms to reduce computational load and enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a comprehensive object detection model processes the entire image to ensure high detection accuracy, then the detection precision is improved, but the computational cost and energy consumption increase significantly
Solution Approach 1:
The patent divides the image processing task into two segments: a first detection step that processes the entire image to identify potential object locations, and a second detection step that processes only selected sub-images at those locations. This segmentation reduces the total computational load and energy consumption while maintaining detection accuracy by focusing detailed analysis only on regions containing objects.
Solution Approach 2:
The patent applies partial action by performing comprehensive detection only on selected sub-images rather than the entire image. The first detection step identifies regions of interest, and the second detection step performs detailed analysis only on those specific sub-images, avoiding unnecessary computation on background areas and thus reducing energy consumption.
2Measurement precision
If a comprehensive object detection model processes the entire image to ensure high detection accuracy, then the detection precision is improved, but the runtime and processing time increase significantly
Solution Approach 1:
The patent segments the detection process into two stages: a fast first detection step that scans the entire image to locate potential objects, and a more detailed second detection step that processes only selected sub-images. This segmentation reduces overall processing time by avoiding comprehensive analysis of the entire image while maintaining high detection accuracy through focused detailed examination of relevant regions.
Solution Approach 2:
The patent implements partial action by performing detailed detection only on selected sub-images identified by the first detection step. This approach reduces processing time significantly compared to comprehensive full-image detection, as computational resources are concentrated only on regions containing objects rather than processing the entire image uniformly.
3Speed
If existing detection models are used to track fast-moving objects, then the tracking capability is provided, but the detection accuracy deteriorates when objects move quickly and erratically
Solution Approach 1:
The patent applies preliminary action by performing the first detection step to identify potential object locations before conducting the second, more accurate detection step on selected sub-images. This preliminary identification allows the system to prepare for tracking fast-moving objects by pre-locating them in the sequence, improving detection accuracy even when objects move quickly or erratically between frames.
Solution Approach 2:
The patent ensures continuity of useful action by applying both detection steps across a sequence of images. The first detection step continuously identifies potential objects in each frame, and the second detection step continuously verifies their presence in selected sub-images, maintaining accurate tracking of fast-moving objects through continuous detection across the image sequence rather than isolated frame-by-frame analysis.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to one aspect of the present disclosure, there is provided a computer-implemented method for object detection, the method comprising: performing a first detection step in a first image of a sequence of images to detect at least one object; determining a second image from the sequence of images wherein the second image is the same first image or temporally follows the first image; selecting a sub-image of the second image in which the at least one object has been detected in the first detection step; and performing a second detection step to detect the at least one object in the selected sub-image.