Harindham/Video_Detections

Each row in the dataset corresponds to one detected object instance in a single frame, along with spatial and temporal metadata.

Detection & Temporal Indexing

The video was sampled at a constant rate (5 seconds per frame).

Each sampled frame is assigned:

• —A frame index (sequential order)

• —A timestamp in seconds indicating its position in the video.

YOLO was applied to each frame to produce:

• —Bounding box coordinates

• —Component class labels

• —Confidence scores

This temporal indexing allows detections to be grouped into continuous time intervals, making it possible to determine when specific vehicle components appear in the video and retrieve all occurrences over time.

Schema

Each detection record includes:

• —video_id – identifier of the source video

• —frame_index – index of the sampled frame

• —timestamp – time position in seconds

• —class_label – detected vehicle component

• —xmin, ymin, xmax, ymax – bounding box coordinates

• —confidence_score – detection confidence from the model

Multiple rows may exist for a single frame if multiple objects are detected.