Aim of the project: In cooperation with LTD “MONDOT”, develop a computationally efficient neural network-based module, which would be able to detect several classes of objects (people, cars, trucks, scooters, etc.) in surveillance videos.
- Explore different novelties in the deep learning field to improve computationally efficient recurrent neural network-based object detection method, developed by EDI leading researcher Roberts Kadiķis (RNN-VDL).
- Develop a specific data annotation tool to speed up the labeling process and create datasets for training deep networks.
- Identify and train accurate and efficient neural network models and fine-tune their hyperparameters.
- Implement the trained models on Raspberry Pi 4 computer, and demonstrate real-time object detection on video data.