研究目的
To introduce BIND-Lite, a real-time texture-less object recognition algorithm capable of executing seamlessly on current mobile devices, by altering various aspects of the original BIND for quicker processing and hardware acceleration.
研究成果
BIND-Lite achieves competitive detection rates in texture-less datasets and integrates seamlessly into mobile devices in real-time, demonstrating the feasibility of real-time texture-less object recognition on mobile platforms.
研究不足
BIND-Lite sacrifices about 10% of its original recognition performance for computational efficiency. The study focuses on Apple devices, limiting generalizability to other platforms.
1：Experimental Design and Method Selection:
The study modifies the BIND algorithm to create BIND-Lite, focusing on reducing computational complexity while retaining detection robustness. Techniques include altering keypoint detection, descriptor design, and employing hardware acceleration.
2：Sample Selection and Data Sources:
The algorithm is tested on the Tools and MOD texture-less datasets.
3：List of Experimental Equipment and Materials:
Intel dual-core i7 Haswell processor with 8GB of memory, Apple devices for mobile implementation.
4：Experimental Procedures and Operational Workflow:
BIND-Lite's performance is compared with SIFT, ORB, BOLD, BORDER, and the original BIND in terms of accuracy and speed. A mobile app, IMPRINT, is developed to showcase BIND-Lite in an AR setting.
5：Data Analysis Methods:
Detection and timing results are consolidated into ROC curves to evaluate performance.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
