A new CNN-BASED object detection system for autonomous mobile robots based on real-world vehicle datasets

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Autonomous mobile robots (AMRs) face navigation challenges in object detection for safe delivery.
• Accurate and efficient object recognition is crucial for AMR navigation systems.

Purpose of the Study:

• To develop a novel convolutional neural network (CNN)-based object detection system for AMRs.
• To improve the accuracy and speed of object identification in diverse real-world conditions.

Main Methods:

• Created original real-world image datasets from Banda Aceh city.
• Developed a CNN-based system using SSD Mobilenetv2 FPN Lite 320 × 320 architecture.
• Retrained the model with custom datasets for object identification (cars, motorcycles, people, rickshaws) under varying light.

Main Results:

• The proposed CNN model demonstrated improved classification and detection accuracy after retraining.
• Quantitative and qualitative evaluations confirmed the model's performance.
• The system effectively identified key objects in diverse lighting conditions.

Conclusions:

• The developed CNN-based object detection system shows significant potential for enhancing AMR navigation.
• The system's ability to perform under varied conditions makes it suitable for real-world applications.
• Further development could integrate this system into practical AMR delivery solutions.