# Issue #36 — Implement `REPEATED_APPROACH` detection

[Devnil434]

Description

A person repeatedly entering and leaving a restricted area is often a strong
suspicious behavior signal.

The current ActionClassifier does not detect repeated zone approaches.

Implement:

ActionHint.REPEATED_APPROACH

inside the action classification pipeline.

---

Expected Behavior

Trigger REPEATED_APPROACH when:

• a person enters a restricted zone
• exits it
• re-enters again
• repeated entry count >= 2

---

Suggested Logic

zone_entry_count = memory_store.get_zone_entry_count(
    obj.track_id,
    zone
)

if zone_entry_count >= 2:
    return ActionHint.REPEATED_APPROACH

---

Challenge

classify_action() currently has no direct access to MemoryStore.

Possible approaches:

1. inject optional store dependency
2. precompute counts in pipeline.py
3. pass counters as parameters

---

Files to Touch

services/memory/action_classifier.py
services/memory/pipeline.py
tests/test_memory.py

---

Acceptance Criteria

• Fires only on second+ entry
• Does not spam every frame
• Supports cooldown behavior
• Existing tests remain green
• Add at least 3 unit tests

---

Helpful Context

Relevant modules:

services/memory/action_classifier.py
services/memory/pipeline.py
libs/schemas/memory.py

[tirth1356]

I’d like to work on this issue. I plan to implement the ActionHint.REPEATED_APPROACH flow inside the action classification pipeline by tracking repeated restricted-zone entries per track_id and triggering the action only on the second+ valid re-entry.

I’ll update the pipeline/classifier integration to provide access to the required entry counters, while ensuring cooldown handling prevents repeated frame-level spam. I’ll also add comprehensive unit tests covering repeated entry detection, cooldown behavior, and non-trigger scenarios while keeping existing tests green.

Please assign this issue to me under gssoc