Custom Object Tracking

Introduction

This repo provides function call to track multi-objects in videos with a given trained object detection model and a source video file as inputs. The tracking approach used in the repo is DeepSort - Simple Online and Realtime Tracking with a Deep Association Metric

Installation

Please install Tensorflow 2 Object Detection API and add the path to your environment.

Usage

1. Clone the github repository

git clone https://github.com/sek788432/Custom_Object_Tracking.git

2. Execution (In two ways)

• saved_model_path: your own object detection model ckpt path
• test_path: test video path
• label_map_path: label_map.pbtxt path
• deep_sort_model: deep sort model path
• output_path: output video path
• min_score_thresh: the minimum score threshold of object detection model

1. Run tracking.py

• e.g.

  cd  Custom_Object_Tracking/objectTracker/
  python tracking.py \
      --saved_model_path=exported-models/ssd_resnet50_119ckpt \
      --test_path=test_video.mp4 \
      --label_map_path=label_map.pbtxt \
      --deep_sort_model=data/mars-small128.pb \
      --output_path=test_video_tracking.mp4 \
      --min_score_thresh=.5
  

2. Call TrackVideo function

• e.g.

  from Custom_Object_Tracking.objectTracker.tracking import TrackVideo
  TrackVideo(label_path, model_path, video_path,
                 output_path, threshold, deep_sort_model)
  

Result

Tracking vehicle by our own model (SSD ResNet50 trained on Waymo Dataset)