10th National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics NCVPRIPG 2025
                   16 -18 July 2025 @ NIT, Srinagar
Human Segmentation, Background Removal, and Super-resolution Enhancement in Residential Surveillance

Advancing Vision Research for a Safer Tomorrow

Challenge Overview
Residential surveillance plays a pivotal role in ensuring the safety and security of smart home environments. The ability to accurately detect, segment, and analyze human presence from video footage is critical for a variety of applications, including intrusion tracking, object detection, and behavior modeling. However, real-world surveillance systems often face significant challenges such as low-resolution footage, dynamic lighting conditions, occlusions, and background clutter, which degrade the performance of existing systems.The Residential Activity Capture Dataset (RACD) has been developed as a benchmark resource to address these challenges and foster research on enhancing human-centric surveillance pipelines. It focuses on human segmentation, background removal, and super-resolution enhancement for improved surveillance in residential settings. This challenge invites participants to create solutions that tackle these core problems and push the boundaries of near real-time video analysis.
Key Objectives
  • Human Segmentation: Accurately detect and segment human figures from residential surveillance footage using bounding boxes.
  • Background Removal: Eliminate all background elements from the segmented human figures, leaving only the human presence.
  • Super-resolution Enhancement: Apply deep learning based super-resolution techniques to enhance the visual clarity of the segmented human regions, improving details such as facial features or clothing.
SpeakerCamera
Significance of the Challenge
This challenge emphasizes human-centric tasks in residential surveillance, encouraging the development of lightweight, accurate, and privacy aware models. By focusing on human segmentation, background removal, and super-resolution enhancement, this challenge directly contributes to the creation of intelligent surveillance systems capable of operating under real-world conditions. It also addresses privacy concerns by eliminating unnecessary background data and ensuring that only human figures are processed for further analysis. Successful solutions from this challenge can help advance applications like, 
  • Identity Verification: Using enhanced, background free human features for biometric recognition.
  • Intruder Detection: Detecting unauthorized persons based on super-resolved human figures.
  • Behavioral Modeling: Understanding and predicting human activities and interactions within a residential setting.

Links

REGISTER HERE

SUBMISSION

Challenge Rules

Eligibility

Open to all: The challenge is open to individuals, teams, academic institutions, and organizations from any background, including researchers, data scientists, engineers, and students. Team Size: Participants can enter individually or as part of a team. Teams are limited to a maximum of 4 members. Age Restriction: There are no age restrictions for participation. Pre-existing Work: Solutions based on pre-existing models are allowed, but participants must ensure they modify and adapt the methods for the specific task and dataset provided.

Registration

Team members may belong to the same or different institutions or organizations. A participant can only be part of one team. Multiple team registrations by the same individual will lead to disqualification. Only one team registration is allowed per team. Duplicate or redundant registrations will not be entertained. There is no restriction on the number of teams from a single organization, as long as the team compositions are unique. Complete your registration using the REGISTER NOW!

Dataset Usage

Confidentiality: The dataset must only be used for the purpose of this challenge. Participants are prohibited from using the dataset for any other research, publication, or external project. No Redistribution: Participants may not share the dataset or its contents with any unauthorized third parties. Dataset Modification: Participants are allowed to preprocess or augment the dataset, but they cannot alter or remove any annotations provided within the dataset. We will release a training dataset of videos and later a testing dataset of videos.

Information

Submission Rules

All challenge submissions must be made through the submission link. Each submission must be a compressed (zipped) file following the naming convention: TeamName_RACD_NCVPRIPG2025.zip The zip archive must contain the following: • A README file with setup instructions, how to run the code, and evaluation notes • The full source code (Python) along with any trained model weights or configuration files • A demo script to reproduce the results • A methodology summary in PDF or DOCX format • Visual results of human segmentation, background removal, and super-resolution • Quantitative results based on the challenge metrics (PSNR, SSIM, MSE, F1 Score, IoU, MAE) • A file listing the inference time and hardware specifications

Evaluation Criteria

Submissions to the RACD Challenge 2025 will be evaluated based on a weighted scoring system that balances super-resolution quality, segmentation accuracy, and model efficiency. The goal is to encourage the development of lightweight, accurate, and deployable models that can be used in real-world surveillance scenarios. Quantitative results based on the metrics (F1 Score, SSIM, PSNR, MSE, IoU, MAE) The number of trainable parameters in the model should ideally be in the range of 100,000 to 5,000,000. This ensures that models with fewer parameters receive a higher score, promoting deployable, edge-compatible solutions.

Important Dates

Web Posting of Challenges: April 22, 2025 Release of Training Dataset: May 1, 2025 Opening Date for Submission to Challenges and release of Test dataset: May 15, 2025 Closing Date for Submission to Challenges: June 5, 2025 Announcement of Challenge Winners: June 25, 2025 Registration Deadline for Challenge Attendees: July 1, 2025

Awards and Recognition
  • The top three teams will be invited for a presentation of their solution in a dedicated session at NCVPRIPG 2025

  • Collaboration on writing summary paper

  • Certificate to each participant

Organizing Members

Prof. Dr. Bindu V R

Professor & Head, School of Computer Sciences, Mahatma Gandhi University

binduvr@mgu.ac.in

Prof. Dr. Deepak Mishra

Professor, Department of Avionics, Indian Institute of Space Science and Technology

deepak.mishra@iist.ac.in

Dr. Abdul Jabbar P

Assistant Professor, Graduate School, Stamford International University

abduljabbar.perumbalath@stamford.edu

Nisha Shamsudin & Mintu Movi

Research Scholars, School of Computer Sciences, Mahatma Gandhi University

nisha@mgu.ac.in, mintu@mgu.ac.in

Contact

  • School of Computer Sciences, Kottayam, Kerala, India
  • Mahatma Gandhi University

Gallery

I BUILT MY SITE FOR FREE USING