Target Detection Network with Multi-Scale Receptive Fields
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target detection methods struggle with robustness in complex scenarios, particularly when targets are occluded, and fail to effectively handle targets of different scales and low-resolution features in natural environments.
Innovation Solution
A target detection method that utilizes a candidate region generation network to generate regions of interest, which are then processed by a target key point network to create a heat map, and subsequently fused with a global feature map to enhance detection accuracy and robustness using local key point information, incorporating convolutional layers with varying receptive fields to handle targets of different scales.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single-scale detector is used, then the device complexity is reduced, but the detection accuracy for targets of different scales deteriorates
Solution Approach 1:
The detector is divided into multiple detectors with different receptive fields (small-scale detector, medium-scale detector, large-scale detector). Each detector is specialized for detecting targets of specific scales, allowing the system to handle multi-scale targets effectively without requiring a single complex detector structure.
Solution Approach 2:
The patent introduces the dimension of receptive field size variation by using dilated convolutions with different dilation rates. This allows the same detector structure to effectively detect targets of different scales by changing the receptive field size, avoiding the need for multiple separate detector structures.
2Device complexity
If manual feature methods are used, then the device complexity is reduced, but the robustness in complex natural scenarios deteriorates
Solution Approach 1:
The patent replaces manual feature extraction methods with deep learning-based automatic feature extraction. Convolutional neural networks automatically learn robust features from data, eliminating the need for hand-crafted features and significantly improving detection robustness in complex natural scenarios.
Solution Approach 2:
The patent uses data augmentation techniques to change the parameters of training data (rotations, flips, scaling, color adjustments). This allows the model to learn more robust features by training on varied versions of the same data, improving generalization to complex natural scenarios without increasing model complexity.
3Measurement precision
If depth feature learning is used, then the detection accuracy is improved, but the detection performance under serious occlusion deteriorates
Solution Approach 1:
The patent segments the target detection task into multiple scales by using detectors with different receptive fields. When a target is occluded, smaller-scale detectors can still detect visible parts of the target, while larger-scale detectors provide contextual information, enabling the system to handle occlusion better while maintaining high detection accuracy.
Solution Approach 2:
The patent merges the detection results from multiple detectors with different receptive fields. By combining the strengths of small-scale detectors (detailed local features) and large-scale detectors (global contextual features), the system achieves both high detection accuracy and improved robustness under occlusion conditions.
Data Source
AI summary
A target detection method and apparatus. The method comprises: acquiring an input image, and sending same to a candidate region generation network to generate a plurality of regions of interest; formatting the plurality of regions of interest, and then sending same to a target key point network to generate a thermodynamic diagram; using a global feature map of the input image to perform convolution on the thermodynamic diagram, so as to generate a local depth feature map; and fusing the global feature map and the local depth feature map, and detecting a target therefrom by means of a detector. The present invention can be applied to target detection at different scales, improves the detection accuracy and robustness of a target detection technique for an occluded target in complex scenarios, and achieves, by means of making full use of local key point information of the target, target positioning under occlusion.