Side-View People Counting With Linked Head-Body Bounding Boxes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional imaging systems struggle to accurately count people from side views due to issues with redundant detection and low performance when individuals overlap, leading to inaccuracies.
Innovation Solution
A surveillance camera system that utilizes bounding boxes for head and body regions to determine accurate counting by intersecting a preset counting line, linking bounding boxes from the same object, and employing virtual bounding boxes for obscured individuals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional image sensors use head detection from top view to count people, then counting accuracy is maintained, but the system cannot be applied to side views and loses versatility
Solution Approach 1:
The patent segments the person detection task into two distinct bounding box types: head region bounding boxes and entire body region bounding boxes. This segmentation allows the system to process side view images by detecting both head and body separately, then using their spatial relationship to confirm person presence and avoid redundant detections, thereby enabling side view applicability while maintaining counting accuracy
Solution Approach 2:
The patent transitions from traditional top-view single-bounding-box detection to side-view dual-bounding-box detection by adding a dimensional aspect. By detecting both head and entire body regions in side views and utilizing their vertical spatial relationship, the system extends people counting capability to side view angles while preserving measurement precision through multi-region verification
2Adaptability or versatility
If image sensors detect people from side views using conventional methods, then view angle flexibility is improved, but detection accuracy deteriorates due to redundant detection and overlapping individuals
Solution Approach 1:
The patent merges head region detection and entire body region detection into a unified counting system. By combining information from both bounding boxes and requiring their spatial relationship to be satisfied, the system achieves more reliable person detection in side views, overcoming the limitations of using either detection method alone and improving detection accuracy despite view angle challenges
Solution Approach 2:
The patent implements a feedback mechanism where the detection results of head bounding boxes and body bounding boxes mutually validate each other. The system checks whether detected head regions have corresponding body regions in expected spatial positions, and vice versa, creating a feedback loop that eliminates redundant detections and confirms actual person presence, thereby maintaining detection accuracy in side views
3Reliability
If the system uses both head bounding boxes and body bounding boxes for counting, then detection reliability is improved, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by first detecting head regions and generating head bounding boxes, then using these as reference points to search for corresponding body regions. This sequential approach with pre-established head detections reduces the overall processing complexity compared to simultaneous multi-object detection, while maintaining high reliability through the two-stage verification process
Data Source
AI summary
A surveillance camera and a control method for the surveillance camera are disclosed. The present disclosure makes it possible to determine whether at least one of a bounding box for a head region of an objected detected from a side view image and a bounding box for an entire body region thereof intersects a preset counting line, thereby counting people passing through the counting line. In the present disclosure, one or more of a surveillance camera, an autonomous vehicle, a user terminal, and a server may be linked to an artificial intelligence module, a robot, an augmented reality (AR) device, a virtual reality (VT) device, and a 5G service.