Two-Level CNN Target Detection in Video Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target detection methods in videos suffer from low accuracy due to the reliance on single-frame image analysis, which fails to effectively capture the features of a target over multiple frames, leading to errors in detecting suspected target regions.
Innovation Solution
A target detection method utilizing a pre-trained Convolutional Neural Network (CNN) model with a two-level classifier system, where the first-level classifier identifies candidate target regions and the second-level classifier refines the detection using feature fusion across multiple frames, improving accuracy by considering the correlation between images in a video.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If single-frame image analysis is used for target detection, then the detection speed is fast and the process is simple, but the detection accuracy is low due to inability to capture target features over multiple frames
Solution Approach 1:
The detection process is segmented into two distinct levels: a first-level classifier for initial candidate region detection and a second-level classifier for refined target detection. This segmentation allows the system to process multiple frames efficiently at the first level while applying more complex analysis only to promising candidates at the second level, thereby improving accuracy without proportionally increasing overall complexity
Solution Approach 2:
The first-level classifier performs preliminary detection to identify candidate target regions before the second-level classifier conducts detailed analysis. This preliminary action filters out obvious non-targets early in the process, allowing the more computationally intensive multi-frame analysis to be applied selectively only to regions with potential targets, thus balancing accuracy improvement with computational efficiency
2Measurement precision
If multi-frame image analysis is used for target detection, then the detection accuracy is improved by capturing target features over multiple frames, but the computational complexity increases
Solution Approach 1:
The system applies multi-frame analysis partially rather than to all frames uniformly. The first-level classifier quickly identifies candidate regions using simplified processing, and only these candidate regions undergo the more computationally demanding second-level multi-frame analysis. This partial application of complex processing significantly reduces overall computational power consumption while maintaining the accuracy benefits of multi-frame analysis where needed
Solution Approach 2:
Different levels of analysis are applied to different regions of the video stream: candidate regions identified by the first-level classifier receive intensive multi-frame local quality analysis at the second level, while non-candidate regions receive minimal processing. This local differentiation ensures high accuracy for potential targets while minimizing computational waste on regions unlikely to contain targets
3Measurement precision
If single-frame detection is used, then the processing time is short and real-time detection is easier, but the detection accuracy suffers due to feature similarity between different actions at single moment
Solution Approach 1:
The detection system operates periodically across multiple frames rather than analyzing each frame in isolation. By examining targets across a sequence of frames with periodic intervals, the system captures temporal patterns that distinguish different actions (e.g., running vs. walking) even when single-frame appearances are similar. This periodic multi-frame analysis improves accuracy while the two-level structure keeps processing time manageable
Data Source
AI summary
Embodiments of the present application disclose a target detection method and device, and relate to the technical field of video processing. The method comprises: obtaining an image sequence to be detected from a video to be detected according to an image sequence determining algorithm based on video timing (S101), extracting a first CNN feature of the image sequence to be detected based on a pre-trained CNN model, performing feature fusion on the first CNN feature based on a second CNN feature to obtain a first fused CNN feature of the image sequence to be detected (S102); inputting the first fused CNN feature into the first-level classifier, and obtaining first candidate target regions of the image sequence to be detected from an output of the first-level classifier (S103); determining a first input region of the second-level classifier based on the first candidate target regions (S104); obtaining a third CNN feature of the first input region based on the first fused CNN feature (S105); inputting the third CNN feature into the second-level classifier, and obtaining a target detection result for the image sequence to be detected based on the output of the second-level classifier (S106).