Evaluating and Improving the Depth Accuracy of Kinect for Windows v2 (2212.13844v1)

Abstract: Microsoft Kinect sensor has been widely used in many applications since the launch of its first version. Recently, Microsoft released a new version of Kinect sensor with improved hardware. However, the accuracy assessment of the sensor remains to be answered. In this paper, we measure the depth accuracy of the newly released Kinect v2 depth sensor, and obtain a cone model to illustrate its accuracy distribution. We then evaluate the variance of the captured depth values by depth entropy. In addition, we propose a trilateration method to improve the depth accuracy with multiple Kinects simultaneously. The experimental results are provided to ascertain the proposed model and method.

• The paper evaluates the depth accuracy of Kinect for Windows v2, models its spatial distribution as a 3D cone, analyzes depth entropy, and proposes a multi-Kinect trilateration method for enhanced accuracy.
• Experimental findings demonstrate optimal accuracy in specific spatial zones but decreased resolution and increased entropy with distance, while the multi-Kinect method successfully improves accuracy compared to single-sensor setups.
• This research provides practical methods to enhance depth data reliability for applications in gaming, VR, and robotics, while also identifying limitations like reflective surfaces for future investigation.

Evaluating and Enhancing the Depth Accuracy of Kinect for Windows v2

This paper investigates the depth accuracy of Microsoft's Kinect for Windows v2 sensor. By addressing a significant gap in understanding the improvements in Kinect v2 in comparison to its predecessor, Kinect v1, the research provides a detailed assessment of the sensor's performance and proposes a method for enhancing depth accuracy using a multi-Kinect setup.

Key Contributions

The authors explore several aspects of depth sensing with Kinect v2:

1. Depth Accuracy Distribution: The paper measures the depth accuracy of Kinect v2 and models this accuracy distribution as a 3D cone. This model serves as a basis for understanding how depth accuracy varies spatially from the sensor's baseline.
2. Depth Entropy Analysis: Entropy, which quantifies the variance in captured depth data, is employed to evaluate the stability and reliability of the depth measurements over time.
3. Multi-Kinect Trilateration: A trilateration method, inspired by its application in GPS and WSN, is proposed to enhance depth accuracy by integrating data from multiple Kinect sensors. This approach calculates optimal depth estimates by solving a system of equations derived from the geometric arrangement of multiple sensors and their respective distance measurements.

Experimental Design and Findings

The paper systematically conducts experiments to validate the accuracy and propose enhancements to Kinect v2:

• Kinect v2 exhibits promising accuracy within specific spatial zones. Depth accuracy is most reliable when objects are positioned within green regions as modeled in the cone distribution, with a demarcation of green (optimal), yellow (acceptable), and red (less reliable) zones of accuracy.
• Depth Resolution and Entropy: The results indicate that depth resolution declines with increasing distance, which is an inherent limitation observed as objects are further removed from the sensor. Additionally, depth entropy increases with distance, suggesting greater variability in repeated measurements at farther ranges.
• Enhancing Accuracy with Trilateration: The multi-Kinect setup demonstrates improved accuracy, particularly when individual Kinect sensors have limitations due to distance or angle of the target. The trilateration approach reduces measurement error and offers an overall robust solution, achieving an acceptable compromise in scenarios where single-sensor setups might fail to deliver reliable data.

Implications and Future Work

This research holds practical implications for fields reliant on affordable 3D sensing technology, such as gaming, virtual reality, and robotics. By effectively utilizing Kinect v2 through enhanced methodologies, applications can benefit from improved depth data reliability and accuracy. Future research could explore adaptive algorithms that dynamically adjust sensor readings based on environmental variables or hardware improvements.

Moreover, while addressing inaccuracies, the paper also highlights limitations such as sensitivity to reflective surfaces and structural noise, which present opportunities for further investigation. Practical enhancements could be realized through sensor fusion techniques, or by calibrating environmental factors impacting signal return.

Through its meticulous exploration of Kinect v2 and contributions to enhancing its utility, this paper lays the groundwork for ongoing innovations in capturing and analyzing depth data with consumer-grade sensors, offering insights into optimizing similar technologies in interactive and immersive environments.