Common issues and human intervention in object detection from handcrafted features to deep learning: discussion.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Traditional object detection relies on handcrafted features and rules.
• Machine/deep learning object detection learns features and rules from data.
• These approaches are often viewed as opposing paradigms.

Purpose of the Study:

• Analyze the commonalities and challenges in object detection workflows.
• Identify areas requiring human supervision in both traditional and machine learning methods.
• Promote a more informed use of object detection techniques and encourage automation research.

Main Methods:

• Comparative analysis of traditional and machine/deep learning object detection.
• Examination of the object detection 'recipe' and its components.
• Review of object detection performance evaluation metrics.

Main Results:

• Object detection, regardless of method, faces three key challenges: object model design, detection robustness, and matching function definition.
• Human supervision is essential for addressing these challenges.
• Current performance metrics also highlight the need for human intervention.

Conclusions:

• Traditional and machine/deep learning object detection share fundamental issues.
• Human intervention remains critical in designing object models, ensuring robustness, defining matching functions, and evaluating performance.
• Further research is needed to automate tasks currently requiring human supervision.