IEEE Signal & Image Processing Challenge Malaysia 2021

(Last updated 19 July 2021)

 

Title: Traffic violation classification

 

* Each team must register through this Google form by 31 July 2021 in order to participate in this challenge.

** At least 1 team member must be an IEEE SPS member

*** test dataset has been released – look for the folder “ExtraDataSet” in the Google drive

 

Background:

In recent years, many dashboard cameras (dash-cam) are installed onto vehicles for video recording. Video footage captured by dash-cam not only records first-hand evidence in case of a car accident for the owners but also serves justice for other victims of road accidents, including hit-and-run incidents, road-bully, reckless drivers, to name a few. On the technical advancement side, this challenge aims to analyze dash-cam video footage for the classification of traffic-violations by developing a fast, robust, and high accuracy algorithm. On the humanitarian side, we aim to create awareness of common road accidents and promote a safer road/environment for commuters to travel in Malaysia and the region.

Data Source
For this challenge, we use the videos published online in a YouTube channel – Malaysian Dash Cam Owner for training and validation purposes.

Original source: Dash camera recording from YouTube channel – Malaysian Dash Cam Owner (https://www.malaysiandashcam.com/)

Link to video compilation: https://youtu.be/RmdhmvIEaGc

Dataset: The dataset, contains the following items:

(i) Videos (21 videos (ID01-ID21) — each video is a compilation of a number of distinct driving scenes

(ii) Annotation master label file (Annotation_master.csv) which contains the labeled occurrences of the traffic violations in each video file, and their associated labels (class, temporal and spatial coordinates). Each occurrence of traffic violation is denoted by a number of objects involved (if there are) followed by a row indicating the temporal segment where the violation had occurred.  For further understanding of the semantic of each column, you may refer to this webpage.

(iii) An image (example.png) showing a representative example of each traffic violation class.

Link: https://drive.google.com/drive/folders/1xjmC6LJy-iRkKeg_81_ShX3CVdzv7FRw     

 

Classes of traffic activity (event)
The following classes will be considered:

ID

Class name

Description

0

sudden_lane_change

A driver/rider in front changes the lane abruptly.

1

illegal_U-turn

A driver/rider making a U-turn while it is not allowed.

2

illegal_crossing

An object illegally crosses the road

3

dangerous_over-take

A vehicle approaching fast from the back and overtaking a vehicle dangerously

4

object_on_road

The vehicle suddenly confronts an object on the road (including animals)

5

wrong_lane

The vehicle is on a lane that it is not supposed to be on (including the reverse drive)

6

other

Other events such as accidents, unexpected behavior, etc.

Evaluation (metric / rubrics)
The following scores will be utilized for evaluation purposes:

Precision = TP / (TP + FP)

Recall = TP / (TP + FN)

F_Score = 2 x  Precision / (Precision + Recall)

A True Positive (TP) denotes an occurrence of traffic violation that is correctly classified (correct class) within an acceptable temporal segment (IoU not less than 0.5*). A False Positive (FP) is an incorrectly detected occurrence of traffic violation that does not exist in the ground truth. A False Negative (FN) is a traffic violation that was missed or not detected.

 * Organizers might use a single threshold or a range of thresholds.

Prize:

Grand prize                 @        RM 500

1st Runner-up             @        RM 300

2nd Runner-up            @        RM 200

*Each participant will receive an electronic certificate of participation.

**All members of the selected teams (submissions) will receive free access to the IEEE ICSIPA 2021 conference technical program and will present their work at the conference.

Timeline (based on MYT: UTC+8):

Release of development data set:      01 June 2021

Registration:                                          31 July 2021

Release of test data set:                     03 August 2021
* Look for the folder “ExtraDataSet” in the same Google drive

Submission deadline:                         10 August 2021

Result announcement                        24 August 2021

Paper submission deadline:               31 August 2021

Award presentation:                            14 Sept. 2021

Registration:

* Each team must register through this Google form by 31 July 2021 in order to participate in this challenge.

* Only registered teams are eligible to take part in the challenge.

* Only one registration is required for each team. Each team consists of at least 1 member and at most 3 members.

At least 1 team member must be an IEEE SPS member.

Submission:

– Submission is done via Google drive by sharing all the following items with the email account (myspoc@ieee.org), AND

Sending an email to myspoc@ieee.org with the title of “Signal & Image Processing Challenge” and include the shared link

– Items required for submission

a. Team name, with member names and emails (please indicate clearly the Team Leader). This can be provided in a text file.

b. Original codes (whole repositories of codes obtained directly from the Internet are strictly not allowed). Only genuine and original work is accepted for consideration. Cite all libraries and sources used.

c. Executable program that runs the codes, with clear instructions provided (e.g.  Readme file containing instructions, or screen captured video) to demonstrate how to execute your codes via the program.

d. The result / output file (.csv) indicating the class of each scene in the video. Note: Program should be able to produce these outputs.

Requirement on output file:

For each test video, the program should first identify a scene, then classify the traffic violation, output the results in the form of a CSV using the following format:

Scene#, start_time_of_scene**, end_time_of_scene**, traffic_event

For example, for the training video #1, the first 3 scenes are as follows:

Scene#

start_time_of_scene

end_time_of_scene

traffic_event

1

0:00

0:13

object_on_road

2

0:14

0:25

illegal_crossing

3

0:26

0:39

wrong_lane

** Time is in seconds.

Paper submission:

Selected teams will be invited to submit a paper to the IEEE ICSIPA 2021 conference for presentation purposes. Papers meeting the technical requirements will be indexed in the IEEE ICSIPA 2021 conference proceeding. For more information about paper submissions, refer to the official website of IEEE ICSIPA 2021: https://sps.ieeemy.org/icsipa2021/

 

Eligibility to participate:

Active SPS membership / IEEE student member at the point of submission

Inquiry:

Please email myspsoc@ieee.org for any questions related to this challenge.