Template Matching for Rapid Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection models require extensive retraining and consume significant memory and processing power, leading to slow image detection and inefficiency in detecting new objects.
Innovation Solution
A method involving an electronic device that generates a template image of a target object, compares it with a detection image through binarization and sliding techniques, and determines candidate regions based on similarity, thereby reducing CPU usage and improving detection speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object detection model is retrained using new training data to detect new objects, then detection accuracy for new objects is improved, but memory consumption increases and detection speed decreases
Solution Approach 1:
The patent segments the object detection task into two parts: using a pre-trained model for general object detection and using template matching for new object detection. This segmentation allows the system to maintain fast detection for known objects while accurately detecting new objects through template comparison, avoiding the need to retrain the entire model and thus maintaining detection speed while improving accuracy for new objects.
Solution Approach 2:
The patent performs preliminary action by pre-training the object detection model on existing data and storing templates of known objects. When new objects need to be detected, the system compares images against these pre-stored templates rather than training from scratch. This preliminary preparation enables fast detection of new objects without the computational overhead of retraining, resolving the contradiction between accuracy and speed.
2Adaptability or versatility
If object detection model is retrained using new training data, then detection capability for new objects is improved, but CPU usage increases
Solution Approach 1:
The patent uses template copying instead of model retraining. Instead of retraining the entire detection model with new data, the system creates templates of new objects and compares images against these templates. This copying approach maintains detection capability for new objects while significantly reducing CPU usage compared to full model retraining, as template matching is computationally lighter than training operations.
Solution Approach 2:
The system performs preliminary template creation for new objects and stores them for later comparison. This preliminary action avoids the need for intensive CPU-based retraining when detecting new objects, as the templates are already prepared and can be quickly matched against incoming images, thus improving adaptability without increasing CPU usage.
3Measurement precision
If large amount of training data is used to train object detection model, then detection accuracy is improved, but memory consumption increases
Solution Approach 1:
The patent extracts only the necessary template information from the training data rather than storing the entire training dataset in memory. By taking out just the essential visual features and creating compact templates, the system achieves high detection accuracy for new objects while dramatically reducing memory consumption compared to storing and processing large amounts of raw training data.
Solution Approach 2:
The system creates simplified copies (templates) of objects from training data rather than storing the original training data. These templates contain the essential visual information needed for detection but occupy much less memory. This copying approach maintains detection accuracy while reducing memory consumption, as the templates are much smaller than the original training datasets.
Data Source
AI summary
A method of recognizing target objects in images obtains a detection image of a target object. A template image is generated according to the target object. The detection image is compared with the template image to obtain a comparison result. Candidate regions of the target object are determined in the detection image according to the comparison result. At least one target region of the target object is obtained from the candidate regions. The method detects target objects in images very rapidly.