Crowd counting Proof of concept (Field deployment)

A snapshot of our crowd counting POC deployed in metro station using two PTZ cameras and a single NVIDIA's Jetson Nano, device, providing real-time counting.

Haris: Stereo-vision based Depth and Size estimation

Haris - The autonomous robot uses stereo-vision based object detection to estimate depth and size dimensions of the object from YOLO bounding boxes to avoid obstacles. This overcomes the limitations of LiDAR based sensing for non-reflective objects such as fire, smoke, etc and detect objects like sign boards. This is the Master thesis work of Layth Hamad . Read the paper in IEEE Sensors Letters..

Lightweight deep learning models

We developed efficient and lightweight model "LCDnet" for crowd density estimation and counting suitable for real-time performance in drones. The performance is validated on NVIDIA's Jetson Nano, and Jetson Xavior devices.

Curriculum learning: Train faster with high accuracy

Curriculum learning (CL) can improve learning and speed up convergence. Want to further reduce the training time of a model? We proposed "CLIP" to reduce training time using CL and dataset pruning.

Crowd counting: Survey paper

New to visual crowd analysis? Read our comprehensive review on crowd counting and visual crowd analysis to get useful insights into benchmark datasets, network architectures, learning methods, model evaluation, and state-of-the-art.

Unauthorized Drone Detection: Survey

Drones can be a threat critical assets, infrastructure, ane privacy of people. A drone detection system shall be efficient, accurate, robust, cost-effective, and scalable. Read our comprehensive review article in IEEE Sensors journal.

Unauthorized Drone Detection: Experiments

We conducted experiment analysis on drone detection methods (i.e., acoustic detection, radio frequency (RF) detection, and visual detection) and our novel encryption-based scheme to detect unauthorized drones in the air. For acoustic detection, we used a ReSpeaker 4-mic array and RaspberryPi processor with Open EmbeddeD Audition System (ODAS) GUI. For RF detection, we used E312 USRP (by Ettusresearch™) with a 2x2 MIMO transceiver. For visual detection, we used a PTZ camera WizSense series (model# SD6CE445XA-HNR) , and Jetson Xavier device.

CODE for IoT

Mobile edge computing (MEC) is promising a lot for 6G and IoT. How efficient edge computing can be? What if the edge resources are insufficient (have you ever heard of "thundering herd problem"?)! We propose a solution CODE to alleviate edge congestion in the event of unplanned traffic spikes.

Distributed Inference in IoT

Video processing using deep learning on the IoT device can be resource consuming, transmitting these can be bandwidth intensive. Distributed inference on local device and edge server can be good if delay deadlines are gauranteed. But how to split the model? An interesting article on distributed inference in Video IoT!

Interested to learn more about edge computing paradigm, read our review article.

ML for Self-Organization Networks (SONs)

Machine learning can play a vital role in building cognitive network architectures that are self-organizing paving way towards zero-touch networks. To enhance end-user experience, transmission throughput is an important metric that has a strong impact on the end-user quality of experience. Our paper provides how to accurately estimate throughput in real-time.

Handover prediction in Wi-Fi

Throughput estimation can be useful but there are other challenges. We proposed ML schemes for prediction of network states (e.g. handover prediction) and efficient decision making (e.g. selection of access point).

Enhanced group formation for Wi-Fi Direct

Wi-Fi Direct, a technology by Wi-Fi Alliance allows Wi-Fi devices to connect to each other without access point (AP). We proposed a vital improvement to the specification that enables optimal dense network for content distribution. Read our work on Enhanced Group Formation in Wi-Fi Direct.

Wi-Fi Direct meets Drones

With enhanced group formation, Wi-Fi Direct enables optimal connectivity in short range communication scenarios e.g., aerial networks. It also offer enhanced power saving for energy efficiency. Our paper proposes Wi-Fi Direct based aerial networking for drones and optimal drones' placement in two real-worls scenarios.

Ramp metering for highways

We propose ramp metering strategy using coordination between upstream traffic signals and the ramp meter. Read the full paper .

Interested in ramp metering, read our review article.