The mission of Kick Analytics is to develop low powered robots that can perform environmental monitoring tasks autonomously. Part of our current project is to develop a surveillance system to detect pedestrians over a large area of land using mobile robots. Few challenges are efficient object detection models that can run real-time tasks on embedded hardware, low light vision, resource constraints. I'm currently working to solve these problems using Machine Learning and Computer VIsion.
• Optimizing object detection models such as YOLOv3 to perform well for real-time applications on small devices.
• 3x improvement in the frame rate compared to the previous implementation on NVIDIA Jetson Nano.
• Developed the documentation for the training pipeline of YOLO on custom datasets.