PTZ Camera Human Detection Using Terrain and HOG Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing PTZ camera systems face challenges in accurately and efficiently detecting human objects due to the trade-off between field of view and zoom, with existing human detection algorithms being insufficiently robust and prone to false positives, especially when background models are incomplete or contaminated by moving nuisances.
Innovation Solution
The method involves creating panoramic terrain and HOG maps to determine typical human size and false detection probabilities, using these maps to refine human detection by converting 2D frame coordinates to 3D world coordinates and applying Histogram of Gradients (HOG) features to reduce false detections and improve detection speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If motion detection or background model is used as a filter for human detection, then the number of false detections is reduced and recognition is speeded up, but the detection becomes less reliable when there is no clean background model or when moving nuisances are present in the scene
Solution Approach 1:
The patent introduces a new dimension to detection by creating panoramic terrain maps and HOG maps that incorporate 3D spatial information and gradient features. Instead of relying solely on 2D motion detection, the system uses terrain maps that encode expected human object sizes at different locations and HOG maps that capture gradient patterns characteristic of human shapes. This dimensional enhancement allows the system to distinguish true human objects from moving nuisances even without a clean background model.
Solution Approach 2:
The patent performs preliminary action by pre-computing panoramic terrain maps and HOG maps during periods when no human objects are present. These maps are created in advance and stored for subsequent detection use. The terrain map pre-computes expected human object sizes at various scene locations, while the HOG map pre-computes gradient patterns. When detection is needed, these pre-computed maps are immediately applied, enabling fast and reliable detection without requiring real-time background modeling.
2Measurement precision
If PTZ camera zooms in to see objects clearly, then object recognition improves, but the field of view decreases and the camera may lose track of moving objects
Solution Approach 1:
The patent applies dynamics by making the PTZ camera's pan, tilt, and zoom parameters adaptive rather than static. The system continuously updates the panoramic terrain map and HOG map based on current camera positions and detected human objects. When a human object is detected, the system dynamically adjusts zoom and pan/tilt to maintain optimal viewing while keeping the object within the field of view. The terrain map is updated to reflect changing camera positions, allowing the system to maintain accurate detection across varying fields of view.
3Measurement precision
If human detection algorithms process all pixels in the frame, then detection accuracy improves, but processing time increases and false detections occur without proper filtering
Solution Approach 1:
The patent segments the detection process into multiple stages using pre-computed maps. Instead of analyzing all pixels uniformly, the system first uses the panoramic terrain map to identify locations where human objects are expected based on pre-computed size information. Then, HOG features are calculated only at these candidate locations using the pre-computed HOG map. This segmentation of the detection process into spatial filtering followed by feature analysis dramatically reduces processing time while maintaining high detection accuracy.
Solution Approach 2:
The patent applies local quality by making detection sensitivity location-dependent through the panoramic terrain map and HOG map. These maps encode different expected human object sizes and gradient patterns for different locations in the scene. The detection algorithm adapts its parameters locally based on the pre-computed map values at each location, rather than applying uniform detection criteria across the entire frame. This location-specific adaptation improves accuracy while reducing false detections.
Data Source
AI summary
The subject matter discloses a method, comprising obtaining a scene from a video camera and defining at least one point of the scene; creating a first scene terrain model of the scene, said first scene terrain model comprises a typical human object size in pixels in said at least one point of the scene; creating a second scene model of the scene, said second model defines a false positive determination that said at least one point comprises a human object; wherein said second scene model is created when the scene does not contain a human object; obtaining an image of the scene, said image is captured by a video camera; determining whether the human object is detected at the at least one point of said captured image by applying said first scene terrain model and said second scene model on the least one point.