A people tracking and counting system based on the low-cost millimeter wave sensor and 3-dimensional Convolutional Neural Networks (3DCNN)

Millimeter wave sensors nowadays play a significant role in the applications of counting and tracking objects due to their great power efficiency and detection performance. Moreover, the technology of object detection has recently obtained important attention because the technologies are beneficial to a variety of fields such as security and energy-efficient applications. In this thesis, we propose a low-cost, and low-complexity but highly accurate method to count and track objects in a small indoor space. The proposed system is developed by applying the low-cost mm-wave radar from Texas Instruments, 3-dimensional RF images which include coordinate and Doppler velocity information, and 3-dimensional Convolutional Neural Networks. To be more specific, the proposed method is able to achieve accuracies as high as 95%, operating at 10 frames per second, to distinguish 4 different moving objects in a small indoor environment.

MAJOR ADVISOR: Thinh Nguyen
COMMITTEE: Jinsub Kim 
COMMITTEE: Bella Bose
COMMITTEE: Raviv Raich
GCR: Roberto Albertani