Vehicle Object Detection Using Color-Infrared Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pedestrian detection systems face challenges in environments with weak illumination, far distances, and occlusion, where existing technologies, such as deep learning architectures for camera and Lidar sensor fusion, require further improvement for accurate and efficient detection.
Innovation Solution
An object detection system utilizing color and infrared cameras, paired with a processor and storage medium, employs a gated fusion unit to combine feature pyramids from Single Shot Detectors (SSDs) for enhanced classification and location data, facilitating robust pedestrian detection in challenging conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional two-stage detectors are used for pedestrian detection, then detection accuracy may be improved, but detection speed and processing time deteriorate
Solution Approach 1:
The system segments the detection task by processing color and infrared image data through separate feature extraction pipelines before fusing them. This allows parallel processing of multiple data sources, improving detection speed while maintaining accuracy through comprehensive feature analysis
Solution Approach 2:
The system merges color and infrared image data at the feature level using a fusion network. By combining complementary information from both modalities, the system achieves high detection accuracy while the efficient feature fusion process maintains fast processing speeds
2Reliability
If deep learning architectures for sensor fusion are implemented, then detection capability in challenging environments improves, but system complexity increases
Solution Approach 1:
The system introduces a fusion network as an intermediary component that receives pre-processed features from both color and infrared pipelines. This mediator integrates the complementary strengths of both sensors, improving reliability in challenging environments while keeping the overall system architecture manageable through modular design
3Measurement precision
If feature fusion is applied to handle multiple scale objects, then detection accuracy improves, but processing time increases
Solution Approach 1:
The system dynamically adapts the feature fusion process based on the scale and characteristics of detected objects. By adjusting the fusion strategy according to object size and detection confidence, the system maintains high accuracy for multiple scale objects while optimizing processing time through selective fusion operations
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An object detection system (20) includes color and infrared cameras (28, 30), a controller-circuit (26), and instructions (40). The color and infrared cameras (28, 30) are configured to output respective color image and infrared image signals (34, 36). The controller-circuit (26) is in communication with the cameras (28, 30), and includes a processor (33) and a storage medium (35). The processor (33) is configured to receive and transform the color image and infrared image signals (34, 36) into classification and location data (41, 43) associated with a detected object (39). The instructions (40) are stored in the at least one storage medium (35) and executed by the at least one processor (33), and are configured to utilize the color image and infrared image signals (34, 36) to form respective first and second maps (50A, 50B). The first map (50A) has a first plurality of layers (54A-64A), and the second map (50B) has a second plurality of layers (54B-64B). Selected layers from each are paired and fused to form a feature pyramid (48) that facilitates formulation of the classification and location data (41, 43).