Research Article
Research on 24-Hour Dense Crowd Counting and Object Detection System Based on Multimodal Image Optimization Feature Fusion
Algorithm 2
Training density map guided detection for RGBD-Net/RGBT-Net.
| Input: Training density map image § = {D1, …, DN}, training | | Epochs Nci, and Adaptive Gaussian kernel initialization input initial mean μ = 0.5 and standard deviation σ = 0.02 | | Onput: A dense crowd detection model with parameters θ AND crowd head detection box | | Initializing density map image D and parameters θ | | for each epoch do | | Step 1: | | If (h == hthermal) | | Gauss kernel matching stop; | | else Continue to match head size according to (5); | | If (h == hDepth) | | Gauss kernel matching stop; | | else Continue to match head size according to (6); | | Step 2: Multivariate Gauss matching with all head sizes; | | Step 3: Multimode features for midterm fusion, according to (9–17); | | Step 4: If (If (8) is true) | | Generating density maps with adaptive Gaussian kernel constraints; | | else Regenerate Gaussian density map according to (7) | | Step 5: Density map guided generation of crowd head detection box; | | Step 6: Update D according to (6). | | end |
|