[Feature]: Add Temporal Action Recognition for Suspicious Activity Detection

[kashviporwal-byte]

@SiddharthRiot
@ArokyaMatthew
@Devnil434

Assign this issue to me under GSSOC'26.

Problem Statement

Currently Eagle performs object detection, tracking, and reasoning effectively, but it lacks temporal understanding of activities across multiple frames.

The system cannot detect actions such as:

• fighting
• running
• loitering
• suspicious movement patterns

Adding temporal action recognition would improve surveillance intelligence and enable activity-level understanding.

Proposed Solution

Implement a lightweight temporal action recognition module using OpenCV + PyTorch.

Suggested implementation:

• Maintain frame-history buffers for tracked persons
• Use temporal models such as CNN+LSTM or MoViNet
• Predict actions from frame sequences
• Integrate action labels into the reasoning pipeline

Example output:
{
"track_id": 12,
"action": "running",
"confidence": 0.93
}

Affected Component

Tracking (ByteTrack / DeepSORT — services/tracking/)

Estimated Difficulty

🔴 Advanced — Spans multiple services or needs ML expertise

Alternatives Considered

Simple frame-by-frame analysis was considered, but temporal models can better understand activities and motion patterns across multiple frames.

Additional Context

Models:

• CNN + LSTM
• MoViNet
• SlowFast

Tech Stack:

• Python
• OpenCV
• PyTorch

Contribution

• I would like to implement this feature and submit a PR.

Checklist

• I have searched existing issues and this is not a duplicate.
• I have read the CONTRIBUTING.md guidelines.

[Devnil434]

proceed now.