Convolutional Neural Network Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection methods, such as R-CNN and Faster R-CNN, are inefficient due to repetitive calculations and lack holistic optimization, leading to prolonged detection times in video monitoring applications.
Innovation Solution
An image processing system utilizing a convolutional neural network that generates feature vectors for objects in images, optimizing the detection process by training the network to determine object positions and categories efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If R-CNN method is used to perform object detection with multiple region proposals and CNN feature calculations, then detection accuracy can be improved, but detection time increases significantly to 2-40 seconds
Solution Approach 1:
The patent merges region proposal generation and CNN feature extraction into a single integrated network that processes the entire image in one pass. The network simultaneously generates region proposals and extracts features for all regions, eliminating the sequential processing of traditional R-CNN and reducing detection time from 2-40 seconds to real-time performance while maintaining detection accuracy.
Solution Approach 2:
The patent performs preliminary feature extraction across the entire image before final object classification. By pre-computing CNN features for all potential regions in parallel during a single forward pass, the system prepares detection data in advance, avoiding repeated calculations and enabling faster final detection decisions.
2Measurement precision
If selective search and SVM classification are used in independent steps for object detection, then detection thoroughness is improved, but the detection process cannot be holistically optimized leading to increased detection time
Solution Approach 1:
The patent combines region proposal generation, feature extraction, and object classification into a single unified neural network. This holistic integration eliminates the independent sequential steps of selective search followed by SVM classification, allowing the system to perform all detection operations in one optimized pass through the network, thereby improving both thoroughness and efficiency simultaneously.
3Measurement precision
If 2000 region proposals are extracted from images with 1-2 seconds extraction time, then detection coverage is improved, but the extraction process consumes significant time affecting overall detection efficiency
Solution Approach 1:
The patent implements continuous feature extraction across the entire image in a single uninterrupted neural network forward pass. Instead of extracting 2000 region proposals sequentially over 1-2 seconds, the network continuously processes the image data flow, generating region proposals and extracting features for all regions simultaneously in one continuous operation, dramatically improving extraction efficiency while maintaining full detection coverage.
Data Source
AI summary
A method configured to implemented on at least one image processing device for detecting objects in images includes obtaining an image including an object. The method also includes generating one or more feature vectors related to the image based on a first convolutional neural network, wherein the one or more feature vectors includes a plurality of parameters. The method further includes determining the position of the object based on at least one of the plurality of parameters. The method still further includes determining a category associated with the object based on at least one the plurality of parameters.