hans-ob1 / deep_eye Goto Github PK

DeepEye is a video surveillance application that runs on your idling desktop computer. This project is inspired from the use cases of object detection using deep learning. There are two modes of detection; motion and objects. The object detection backend uses SSD architecture and detects presence of human and pets. The code base is open-sourced, feel free to modify for your own.

Screenshot 1	Screenshot 2	Screenshot 3

Screenshot 4	Screenshot 5	Screenshot 6

• Ubuntu 16.04.3
• OpenCV 3.1
• Tensorflow r1.3
• Python 3.5
• pyQT5
• virtualenv (recommended)

• Requires a CUDA ready GPU, Cuda driver and cuDNN based requirement of Tensorflow
• The frozen model in models/ssd_mobilenet_v1_model is trained on modified coco dataset, consist of human, cat and dogs.

After you get the dependencies, run the following cmd

python3 interface.py

You can perform your own training using google object detection api. The following command is useful for training your own model and export to frozen model.

python3 train.py \
         --logtostderr \
         --train_dir=training/ \
         --pipeline_config_path=training/ssd_mobilenet_v1_pets.config

For exporting to frozen model

python3 export_inference_graph.py \
        --input_type image_tensor \
        --pipeline_config_path training/ssd_mobilenet_v1_coco.config \
        --trained_checkpoint_prefix training/model.ckpt-490468 \
        --output_directory ssd_3class