Adaptive Gaze Control for Object Detection.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Object detection is crucial for computer vision tasks.
• Existing methods like window-sliding are computationally intensive.
• Gaze-control models offer potential for efficiency but require further development.

Purpose of the Study:

• To introduce a novel, computationally efficient gaze-control model for object detection.
• To enhance object detection performance by leveraging local image samples more effectively.
• To demonstrate the model's capability in face detection tasks.

Main Methods:

• Developed the act-detect model utilizing local image samples for gaze shifting.
• Implemented a novel approach where local samples determine object direction and distance.
• Simultaneously adapted visual features and gaze-control strategy for contextual exploitation.

Main Results:

• act-detect demonstrated high computational efficiency, requiring significantly fewer samples than window-sliding methods.
• The model achieved comparable, and slightly superior, detection performance in face detection tasks.
• The model effectively utilized spatial relations and object context at various scales.

Conclusions:

• The act-detect model presents a computationally efficient and effective alternative for object detection.
• Its advanced use of local samples and gaze control offers improved performance and speed.
• This approach opens new avenues for exploiting object context in computer vision.