A Novel 3D Expansion and Corrosion Method for Human Detection Based on Depth Information.

The existing body detection methods based on depth images mostly depend on the extraction of image gradient features, which is the evolution of the traditional 2D plane image processing method for human body detection. Although their detection accuracy is high, the algorithms consume a large amount of computing and storage resources. Aiming at the real-time demand of safe-driving of forklift trucks in industry, this paper presents a novel 3D expansion and corrosion method for human detection by using depth information. A depth image of human body is detected based on the characteristics of human Head-Shoulder-Body Density (HSBD), which can reduce the error and loss of the depth information caused by changing light conditions, complex background scenes and various distances from objects. Experimental results show that the recognition rate of the proposed method is over 96%, and the recognition speed is over 15 frames per second. This can satisfy the safe-driving demands of forklift truckers in factory.