This projects implement Object Detection based on Deep Learning by using the TensorFlow Object Detection API. It shows good performance of Object Detection models for inference, on Pedestrian Detection.

Refer to the instructions from this link. This tutorial includes TensorFlow installation, environment setup, TensorFlow models installation and protobuf installation/compilation. To annotate training datasets, we should also need LabelImg installation. I would suggest to ceate virtual environment under Anaconda though it is not a requirement.

We upload the test vedio standing1.MP4 in test_vedio folder. And the output of the procesed vedio will locate at the same directory. We will extract frames from the vedio and put them into test_images directory. Then we implement the detection on every frame and the processed images would be saved. Finally we convert all processed images to a vedio.

Tensorflow detection provides a collection of detection models pre-trained on the COCO dataset, the Kitti dataset, the Open Images dataset, the AVA v2.1 dataset and the iNaturalist Species Detection Dataset. We use these pre-trained models to make prediction on our test datasets. we can download the pre-trained models from http://download.tensorflow.org/models/object_detection/.I choose faster_rcnn_resnet50_coco_2018_01_28 as default model which has the best performance on our datasets that I have ever tried.

Run pedestrian_tracking.ipynb and follow the instructions inside.