Object Detection via Neural Network Feature Concatenation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object tracking methods in electric devices rely on predicting the location of a target object using bounding boxes, which can be inefficient for accurate detection, especially when classifying objects with varying confidence levels.
Innovation Solution
An object detecting device and method that utilizes a classifier to generate color and grayscale images, process them through a neural network to create characteristic patterns, adjust dimensions, concatenate them to calculate class confidence, and output classification results if above a preset threshold, facilitating efficient object classification and tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If bounding box prediction is used for object tracking, then object location can be predicted, but detection accuracy and classification reliability deteriorate
Solution Approach 1:
The patent changes the parameter from simple bounding box coordinates to multi-dimensional feature vectors that include color information, gray scale information, and neural network-derived characteristics. This parameter transformation enables both accurate detection and reliable classification by capturing richer object attributes.
Solution Approach 2:
The patent combines multiple types of data (color images, gray scale images, and neural network characteristic patterns) into a composite feature representation. This composite approach allows the system to leverage the strengths of each data type while mitigating their individual limitations, thereby improving both detection accuracy and classification reliability.
2Measurement precision
If neural network algorithms process color and gray scale images, then classification precision improves, but processing time increases
Solution Approach 1:
The patent segments the image processing task into distinct stages: extracting color information, extracting gray scale information, and generating neural network characteristic patterns separately. This segmentation allows each component to be optimized independently and facilitates parallel processing, reducing overall processing time while maintaining high classification precision.
Solution Approach 2:
The patent performs preliminary feature extraction and characteristic pattern generation before final classification. By preparing and pre-processing the data structures (color and gray scale images) in advance, the system reduces the computational burden during the critical classification decision-making process, thereby decreasing total processing time.
3Reliability
If confidence threshold filtering is applied, then false positive reduction improves, but detection completeness may worsen
Solution Approach 1:
The patent implements a feedback mechanism where the confidence threshold is dynamically adjusted based on the distribution of classification results. The system monitors detection performance and adjusts the threshold accordingly, allowing it to adapt to different scenarios and maintain both high reliability (false positive reduction) and detection completeness.
Data Source
AI summary
An object detecting device, an object detecting method and a non-transitory computer-readable medium are provided. The object detecting method includes the following steps: A classifier generates a current color image and a current gray scale image. The classifier generates an initial characteristic pattern from the current color image via a neural network algorithm. The classifier adjusts a current dimension of the initial characteristic pattern to generate an adjusted characteristic pattern according to a gray scale image dimension of the current gray scale image. The classifier concatenates the adjusted characteristic pattern and the current gray scale image to calculate a class confidence. The classifier determines whether the class confidence is larger than a confidence threshold, and outputs a current classification result if the class confidence is larger than the confidence threshold. A storage device stories the current classification result.