Related Experiment Video
Updated: May 10, 2025

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
Acceleration of Image Classification and Object Tracking by the Intel Neural Compute Stick 2 with Power Efficiency
Tianyu Gao1, Jozsef Suto1,2
1Department of Informatics Systems and Networks, Faculty of Informatics, University of Debrecen, Kassai Street 26, 4028 Debrecen, Hungary.
The Intel Neural Compute Stick 2 (NCS2) significantly accelerates deep learning on Raspberry Pi 4B, boosting image recognition by 400% and object tracking by up to 1400%. This enables efficient real-time AI in embedded systems.
Area of Science:
- Computer Vision
- Embedded Systems AI
- Hardware Acceleration
Background:
- Real-time operation of complex neural networks is crucial for embedded systems.
- Deploying deep learning in industrial applications requires cost-effective solutions.
- The Intel Neural Compute Stick 2 (NCS2) offers potential for edge AI acceleration.
Purpose of the Study:
- To evaluate the efficiency and power consumption of the NCS2 on a Raspberry Pi 4B.
- To assess the NCS2's performance in accelerating image classification and object tracking.
- To supplement OpenVINO™ documentation with real-world application data.
Main Methods:
- Utilized Raspberry Pi 4B platform with Intel NCS2.
- Conducted image recognition tests using a single model.
- Performed real-time object tracking tests with two models, employing the Deep SORT algorithm.
- Evaluated performance variations with different model numbers and conditions.
Main Results:
- NCS2 increased image recognition performance by approximately 400%.
- Real-time object tracking performance improved by 1200% to 1400%.
- Achieved over 50 FPS for image recognition and over 20 FPS for object tracking.
- Observed 200% to 400% power efficiency gains with NCS2.
Conclusions:
- The NCS2 provides substantial performance enhancements for AI tasks on constrained hardware.
- The combination of Raspberry Pi 4B and NCS2 is effective for real-time embedded AI applications.
- Significant power efficiency improvements make NCS2 a viable solution for edge computing.
Related Concept Videos
Force Classification
Contact and non-contact forces are two of the most widely used categories of forces. As the name suggests, contact forces require physical contact between two objects to act upon each other. Examples of contact forces include frictional,...
Acceleration Vectors
Average Acceleration
Relative Motion Analysis - Acceleration
PI Controller: Design
Relative Motion Analysis using Rotating Axes - Acceleration
Time differentiation is...

