Dynamic Image Pyramid Construction for Target Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing method of building image pyramids for target object detection is inefficient, particularly in real-time video processing and embedded systems, as it requires significant time and resources, leading to reduced processing speed due to the 'building before using' strategy, where all scales are built statically before detection.
Innovation Solution
A dynamic method for building and detecting target objects in images, where the image pyramid is constructed and features are extracted layer by layer, stopping when a target is detected, and moving to higher resolution layers only if necessary, using bilinear interpolation and feature extraction techniques like luminance, chroma, and histogram of oriented gradients, with the option to down-sample or up-sample intermediate layers based on a 6-power polynomial function.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image pyramids are built before using with all scales statically constructed, then complete multi-scale image information is available for detection, but processing time increases significantly and detection speed decreases
Solution Approach 1:
The patent transforms the static image pyramid construction into a dynamic process where layers are built on-demand based on detection needs. The system starts with a coarse layer and progressively builds finer layers only when targets are not detected, making the data structure construction adaptive and dynamic rather than predetermined and static.
Solution Approach 2:
The patent performs preliminary detection on coarse layers before building finer detail layers. By conducting initial target detection on lower-resolution images first, the system can identify obvious targets early and avoid the time-consuming process of building complete high-resolution pyramids, thus preparing and acting in advance based on preliminary results.
Solution Approach 3:
The patent implements partial action by constructing only the necessary portion of the image pyramid required for detection. Instead of building all layers from the beginning, the system builds layers progressively and stops when detection is successful, performing only the minimal necessary work rather than complete pre-construction of all scales.
2Measurement precision
If dense scale space with high resolution layers is constructed, then detection precision for small objects is improved, but data structure complexity and memory consumption increase
Solution Approach 1:
The patent segments the image pyramid construction into discrete, manageable layers that are built progressively. Instead of constructing the entire dense scale space at once, the system divides the process into separate layer construction steps, building only the current necessary layer and moving to the next layer only when needed, thus reducing overall data structure complexity.
Solution Approach 2:
The patent applies local quality by constructing high-resolution layers only in regions or at stages where detection precision is actually needed. The system builds finer detail layers selectively based on detection results from coarser layers, rather than uniformly constructing all high-resolution layers throughout the entire scale space, thus optimizing memory usage and reducing complexity.
3Adaptability or versatility
If all image pyramid layers are built in advance, then detection can be performed at any scale, but computational resources and processing time are wasted on unnecessary layers
Solution Approach 1:
The patent performs partial action by building only the necessary number of pyramid layers required for successful detection. The system constructs layers progressively and terminates the construction process as soon as a target is detected, avoiding the waste of computational resources on building additional layers that would not contribute to the detection task, thus improving detection efficiency.
Solution Approach 2:
The patent implements self-service where the detection process itself determines when to stop building layers. The system automatically adjusts the construction process based on real-time detection feedback, stopping layer construction when detection succeeds without requiring external intervention or pre-defined limits, thus optimizing resource usage while maintaining flexibility.
Data Source
AI summary
A method for detecting a target object in an image is described. The method includes obtaining an image to be detected, determining a first layer of an image pyramid as a current layer, building an image feature data set of the current layer according to the image to be detected, and detecting the target object therefrom. In response to detecting the target object, the method includes stopping the detection and outputting the detected target object; in response to not detecting the target object and the number of layers of the image pyramid not reaching a threshold for the number of layers, determining a next layer of the current layer as the current layer and returning to the building the image feature data set of the current layer according to the image to be detected and detecting the target object therefrom.