We're developing video analysis solution to easily find out specific objects using informations such as direction, speed, colors, an so on.

Entire project consists of 3 Modules :

• Detection Engine
• Tracking Engine
• Main GUI program

1,2) Detection & Tracking Summary

1-1) Detection: Introduction

1-2) Detection: Algorithm Comparsion

1-3) Detection: Data Processing

1-4) Detection: Model Structure

1-5) Detection: Dataset

2-1) Tracking: Introduction

2-2) Tracking: Tracklet

2-3) Tracking: Trajectory

3) Overlay Summary

Detection engine detects people and cars in each frame.

We use a deep learning algorithm to detect objects. Our CNN model is described as follows:

Now, we can see how RCNet looks a target image.

This is a processed image of the first convolutional layer.

This is a second convolutional layer output image. More brighter, more valuable to see in RCNet.

conv3

The last layer output. Each note include many dimensions of infomation. The role of merge these infomation is Detection Layer as follows:

RapidCheck Detection Engine Result Example: ] Left : General VOC Data Learned model Right : RCNet Trained on CCTV Domain

• tensorflow >= 1.0
• keras >= 2.0
• opencv-python >= 3.0
• matplotlib
• pymysql

Tracking engine implements tracking algorithms and analyzes object's informations as the following pipeline.

1. Read detection responses from database
2. Build Tracklet for short term period
3. Build Trajectory for entire period
4. Extract each object's information such as direction, speed, color

• opencv(c++) >= 3.0
• mysql

This project is based on Visual Studio 2013. Our dependecies are set in cv_x64_debug.props.

We followed Kusmawan's Building and Installing OpenCV tutorial.

Given analysis results, original video is compressed into a short time video using overlay algorithm. User can choose class(person, car), direction, color to find a specific object.

• OxyPlot 1.0
• MaterialSkin 1.0
• MySql 6.9.9
• CefSharp 47.0
• Accord.Video.FFMPEG 3.3
• Accord.Video.VFW 3.4.2
• Accord.MachineLearning 3.5

[1] Zamir AR, Dehghan A, Shah M (2012) Gmcp-tracker: Global multi-object tracking using generalized minimum clique graphs.

[2] Wenhan Luo, Junliang Xing (2014) Multiple Object Tracking: A Literature Review, arXiv:1409.7618

[3] Dehghan A, Shah M , Mubarak S (2015) GMMCP Tracker: Globally Optimal Generalized Maximum Multi Clique Problem for Multiple Object Tracking

[4] Joseph Redmon, Santosh Divvala (2015), You Only Look Once: Unified, Real-Time Object Detection, https://pjreddie.com/darknet/yolo/