Tiny-YOLO object detection supplemented with geometrical data

TL;DR

This paper enhances Tiny-YOLO object detection by incorporating scene geometry and spatial scale information, improving accuracy for autonomous robots with minimal additional computation.

Contribution

It introduces a method to integrate scene geometry and spatial scale data into Tiny-YOLO, improving detection precision in robotic applications.

Findings

Detection with scale channel outperforms standard RGB detection.

Method maintains low computational overhead.

Improves mAP in scene-specific object detection.

Abstract

We propose a method of improving detection precision (mAP) with the help of the prior knowledge about the scene geometry: we assume the scene to be a plane with objects placed on it. We focus our attention on autonomous robots, so given the robot's dimensions and the inclination angles of the camera, it is possible to predict the spatial scale for each pixel of the input frame. With slightly modified YOLOv3-tiny we demonstrate that the detection supplemented by the scale channel, further referred as S, outperforms standard RGB-based detection with small computational overhead.