High Precision Recognition of Human Body from Top View

In order to address the challenge of weak camera-based object detection under the industrial top view angle, particularly when the monitoring angle is from a top-down perspective, where targets are difficult to recognize, this paper proposes a yolov5 object detection algorithm based on CBMA. The algorithm focuses on enhancing the key features of the target area to improve the model’s accuracy. Additionally, the algorithm optimizes the multi-scale feature fusion module and introduces a detection layer specifically designed for small target features at the prediction stage, enhancing the model’s generalization capability. The border regression is optimized using the EIOU (Enhanced Intersection over Union) metric to accelerate the model training convergence speed and improve the accuracy of target positioning. Furthermore, targeted training is conducted using a self-built dataset to further optimize and enhance the model’s performance. Finally, a security system based on an electronic fence is developed using the proposed algorithm.