Executive Summary : | The proposal aims to enhance surveillance solutions by recognizing complex activities and improving object recognition accuracy in low visibility conditions. This involves using mmWave Radar technology to capture a heatmap from reflections of nearby objects, developing a machine learning solution to train the model with heatmaps obtained from different objects, and combining video images with Radar images for better performance. The computation task will be mapped in a distributed edge computing infrastructure consisting of multiple edge devices and edge controllers. The project will consider metrics such as accuracy, latency, wireless bandwidth usage, power consumption, and processing capability/speed. A multi-party access control mechanism will be developed to manage different edge controllers. Experiments will involve preparing labelled datasets corresponding to various activities, using object detection, pose estimation schemes, and developing logical reasoning to predict their activities. Raspberry PI will be used to build mmWave transceivers and collect videos in low light scenarios to test the accuracy of combined video and Radar images. A small prototype will be built in a lab/departmental building, consisting of multiple edge devices and Radars. If these fundamental objectives are achieved, future possibilities include building smart cameras capable of suspicious activity detection, tilting/zooming into specific areas, and improving camera visibility in low light situations. These solutions can also be integrated with cars for safer driving during fog and autonomous driving scenarios. |