.png){kind=icon}
.png){kind=icon}
Return
HARCE: human activity recognition in complex environments an improved YOLOv8s model with Coyote and Badger Optimization for enhanced detection accuracy
.png){kind=icon}
PRE
AI
M
B
I
DOI：10.1007/s13198-026-03205-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The need for enhanced security shows the significance of research in dynamic human activity recognition (HAR). The shortcomings of current frameworks are high computational demands and poor performance. An optimized YOLOv8s model is proposed to detect human activity with increasing accuracy and efficiency. It showcases the following components (1) The Depth-wise Separable Convolution (DSConv) produces multiple feature maps to reduce the computational complexity. (2) The dual-path attention gate (DPAG), to improve the detection accuracy under difficult circumstances. (3) The Feature Enhancement Module (FEM) preserves the target details, important features and detection accuracy. The new metaheuristic algorithm Coyote and Badger Optimisation (CBO) is presented for optimisation. This model is evaluated with the standard datasets (KTH, WEIZMAN, WVU, and IXMAX) and a newly created dataset Group-Behavior-Analysis (GBA) with variety of activities performed by people in natural settings. The proposed model achieves high accuracy even in intricate, real-world video surveillance scenarios.
Keywords:
Human activity recognition
Depthwise separable convolution
Coyote and Badger Optimization
Dual-path attention gate module
Feature extraction
Computational complexity
Journal
I
IF:
1.4
Papers:
291
Citations:
3.0K