Detection and Classification System

A web based implementation of the TensorFlow Object API to detect and classify objects. The inputs can be image, video or real-time web cam feed. The outputs include classified object using bounding boxes.

References

• https://github.com/GoogleCloudPlatform/tensorflow-object-detection-example
• https://github.com/datitran/object_detector_app

Language

• Python 3.5.2

Prerequisites (Tested on Ubuntu and MacOS)

1. TensorFlow 1.3
2. TensorFlow Object Detection API
3. OpenCV 3.2
4. Flask 0.12.2

Install packages required for tensorFlow object detection API and Flask

1. TensorFLow Installation - https://www.tensorflow.org/install/
2. TensorFLow Object Detection API - Refer below
3. OpenCV - https://milq.github.io/install-opencv-ubuntu-debian/
4. Flask

# apt-get update
# pip install Flask==0.12.2 WTForms==2.1 Flask_WTF==0.14.2 Werkzeug==0.12.2

Install the Object Detection API library

# apt-get update
# apt-get install -y protobuf-compiler python-pil python-lxml python-pip python-dev git
# git clone https://github.com/tensorflow/models
# cd models/research
# protoc object_detection/protos/*.proto --python_out=.

Set MODEL_BASE in detector_app.py to the path of the object detection api ../models/research

Download the pretrained model binaries

Models available at https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

There are five pretrained models that can be used by the application. They have diffrent characteristics in terms of accuracy and speed. You can change the model used by the application by setting the PATH_TO_CKPT to point the frozen weights of the required model.

You specify one of the following models.

• ssd_mobilenet_v1_coco_11_06_2017
• ssd_inception_v2_coco_11_06_2017
• rfcn_resnet101_coco_11_06_2017
• faster_rcnn_resnet101_coco_11_06_2017
• faster_rcnn_inception_resnet_v2_atrous_coco_11_06_2017

Running the application

# git clone https://github.com/anoopjeerige/realtimeCV
# cd ~/realtimeCV/obj_detect_multi
# export FLASK_APP=detector.app
# flask run

You have to wait around 60secs for the application to finish loading the pretrained model graph. You'll see the message Running on http://127.0.0.1:5000/ (Press CTRL+C to quit) when it's ready.

Now you can access the instance's IP address using a web browser. When you upload an image file with a jpeg, jpg, or png extension, the application shows the result of the object detection inference. The inference may take up to 10 seconds, depending on the image.

You can check the objects by clicking labels shown to the right of the image.

You can upload an video file (preferably small < 5MB, .mp4, 3gp), for the object detection inference. The Real-time detection take the web cam as the source for the object detection.

The webcam source is default 0, and can be changed depending on the source as defined on the system where the application is running.