Resource-aware video streaming (RAViS) framework for object detection system using deep learning algorithm

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Video stream mining demands significant computational resources, often exceeding the capabilities of limited hardware.
• Analyzing continuous video data streams can lead to system stalls and increased operational costs due to resource limitations.

Purpose of the Study:

• To develop a Resource-Aware Video Streaming (RAViS) framework for efficient object detection on resource-constrained devices.
• To adapt deep learning-based object detection systems, specifically YOLO, to the available CPU, RAM, and storage of a Raspberry Pi.

Main Methods:

• Developed the RAViS framework to monitor and adapt object detection system parameters based on real-time resource availability.
• Utilized video streaming simulations to test the framework's performance in recognizing people in a room using a deep learning model.
• Implemented adaptive strategies to optimize resource utilization for continuous video processing.

Main Results:

• The RAViS framework successfully adapted the YOLO object detection system to the Raspberry Pi's limited resources.
• Experimental results demonstrated that the framework maintained detection accuracy while operating within resource constraints.
• The system ensured continuous operation of the object detection task on the limited hardware.

Conclusions:

• The RAViS framework enables object detection systems to run effectively on computers with limited resources.
• Continuous monitoring and feedback mechanisms allow for dynamic adjustment of detection parameters, optimizing resource utilization.
• This approach ensures the sustained accuracy and effectiveness of deep learning-based object detection in streamed video analysis.