Scribd Logo
Upload

Welcome to Scribd!

  • AI City Challenge 2024

    Uploaded by

    Hridya N
    9 views2 pages

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Flag for inappropriate content
    Download as pdf or txt
    9 views2 pages

    AI City Challenge 2024

    Uploaded by

    Hridya N

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Flag for inappropriate content
    Download as pdf or txt
    of 2

    AI City Challenge 2024

    Group Members

    Hridya N - TVE23ECRA08

    MUHAMMED ASLAM - TVE23ECRA11

    MUHAMMED RISWAN – TVE23ECRA13

    1. Project Title: Traffic Safety Description and Analysis

    2. Aim:

    • Develop a machine learning model capable of generating dense video captioning captions for
    traffic safety scenarios, focusing on pedestrian accidents, using long videos with multiple
    viewpoints.
    • Leverage multiple cameras and viewpoints to describe the continuous moment before incidents,
    as well as normal scenes, capturing details regarding context, attention, location, and behavior of
    pedestrians and vehicles. participants will detail events leading up to incidents and ordinary
    scenes, offering deep insights for applications like insurance inspection processes and accident
    prevention.

    3. Deliverables:

    • Trained machine learning model capable of fine-grained video captioning for traffic safety
    scenarios.
    • Detailed documentation outlining the model architecture, training procedure, hyperparameters,
    and evaluation results.
    • Codebase implementing the model, preprocessing pipeline, and evaluation metrics.
    • Evaluation report showcasing the performance of the model on the provided dataset and
    comparison with baseline approaches.

    4. Brief Review of Papers in the Related Area:

    Paper 1: Deep Visual-Semantic Alignments for Generating Image Descriptions (2015)

    • Details: Proposes an approach using convolutional neural networks (CNNs) and recurrent neural
    networks (RNNs) with attention mechanisms for image captioning.
    • Pros: Effective in generating descriptive captions by aligning visual and semantic information.
    • Cons: May struggle with capturing fine-grained details in complex scenes with multiple objects
    and interactions.
    Paper 2: Attention Mechanisms in Neural Networks for Video Captioning (2017)

    • Details: Introduces attention mechanisms for video captioning tasks, allowing the model to focus
    on relevant regions of the video frame.
    • Pros: Improves the model's ability to generate informative captions by attending to salient
    features.
    • Cons: Increased computational complexity and potential performance degradation on long
    videos with multiple viewpoints.

    Paper 3: Transformer-based Models for Video Captioning (2022)

    • Details: Explores the adaptation of transformer-based architectures, originally designed for


    natural language processing tasks, to video captioning.
    • Pros: Enables efficient modeling of long-range dependencies and captures temporal dynamics in
    videos effectively.
    • Cons: Requires large amounts of training data and computational resources for training, limiting
    applicability in resource-constrained environments.

    5. Research Gap:

    • While existing research has made significant progress in video captioning, there is a need for
    specialized models tailored to fine-grained traffic safety scenarios.
    • Current approaches may not adequately capture the nuanced details of pedestrian accidents and
    the surrounding context, highlighting the importance of developing dedicated solutions for this
    domain.

    You might also like