Image Sequence Detection Using Historical Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image sequence detection methods rely solely on current data features, neglecting historical data features, which results in low accuracy of classification algorithms.
Innovation Solution
An image sequence detection method that incorporates both historical and current data features by using a pretrained target detection model, where the initial image sequence is inputted to obtain a detection result sequence, and the detection results are determined for each image in the sequence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only current data feature is used for classification, then the classification process is simple, but the image sequence detection accuracy is low
Solution Approach 1:
The patent applies preliminary action by pre-training a target detection model using historical data features before actual detection. The model is trained offline with sample training data that includes historical features from continuous frames, so that during runtime, the model can directly utilize these pre-processed historical features without adding real-time computational complexity. This resolves the contradiction by preparing the classification system in advance to handle both current and historical features accurately.
Solution Approach 2:
The patent introduces historical data features as an additional dimension to the classification process. Instead of only using current frame features, the system incorporates features from previous frames (i-1)th frame alongside the current (i)th frame, creating a multi-dimensional feature space. This dimensional expansion enables more accurate detection by considering temporal context without fundamentally complicating the classification architecture.
2Measurement precision
If historical data feature is incorporated into classification, then detection accuracy improves, but computational load increases
Solution Approach 1:
The patent reduces real-time computational energy consumption by performing feature extraction and model training in advance. Historical data features are extracted and stored during the pre-training phase, and the target detection model learns to process these features offline. During actual image sequence detection, the system only needs to retrieve pre-extracted historical features and perform forward propagation through the trained model, significantly reducing real-time computational energy requirements while maintaining high classification accuracy.
3Measurement precision
If only current frame is used for detection, then processing speed is fast, but detection accuracy is low
Solution Approach 1:
The patent uses copying by storing and reusing historical data features from previous frames as reference information. Instead of reprocessing all historical data, the system copies relevant feature representations from the (i-1)th frame and combines them with the current (i)th frame features. This approach enables the model to leverage temporal context for improved accuracy while maintaining fast processing speed, as the copied features are pre-extracted and ready for immediate use in classification.
Data Source
AI summary
This application discloses an image sequence detection method performed by an electronic device. The method includes: obtaining an initial image sequence, the initial image sequence including a target object; inputting the initial image sequence to a pretrained target detection model, to obtain a detection result sequence, the target detection model being obtained by inputting sample training data to an initial detection model to be trained for training, the sample training data including one set of sample feature information respectively extracted from one set of sample images, and each piece of sample feature information including feature information jointly determined by using one frame of sample image and a previous frame of sample image of the one frame of sample image; and determining a detection result corresponding to each image in the initial image sequence based on the detection result sequence, the detection result indicating a category of the target object.