研究目的
To develop an accurate, area-efficient, and ultra-low power real-time gesture recognition system for smart wearable devices, addressing the limitations of existing systems in power consumption and size.
研究成果
The proposed gesture recognition system achieves high accuracy (90.6%), ultra-low power consumption (137 pW), and small area (1.78 mm2) by using a memory-efficient peak-based classification engine and low-power on-chip image sensor. It is suitable for IoT and wearable applications where power and area constraints are critical, outperforming existing systems in these metrics.
研究不足
The system is designed for specific hand gestures (8 types) and may not generalize to all gesture variations; the low-resolution image sensor (32x32 pixels) might limit detail capture; performance is optimized for ultra-low power and small area, potentially at the cost of higher complexity gestures.
1：Experimental Design and Method Selection:
The system integrates an on-chip 32x32 image sensor with a digital signal processor (DSP) featuring a peak-based gesture classification engine (PGCE), motion detection unit, and top controller. It uses a 3D feature vector based on row-column peaks for gesture classification, with fixed-pattern noise elimination and parallelism for efficiency.
2：Sample Selection and Data Sources:
Hand gestures are captured using the on-chip image sensor, with data processed in real-time for recognition.
3：List of Experimental Equipment and Materials:
A test chip fabricated in 65nm CMOS technology, including the image sensor, DSP, and on-chip memory.
4：Experimental Procedures and Operational Workflow:
The image sensor captures frames; motion detection triggers saving of two consecutive frames; feature extraction cores compute row and column sums, detect peaks, and classify gestures using the 3D feature vector; parallelism reduces latency.
5：Data Analysis Methods:
Recognition accuracy is measured, power consumption is monitored at various supply voltages, and comparisons are made with existing systems using performance metrics.
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
