Abstract:
Automatic recognition of people's walking patterns and the development of identification methods have enabled new developments in the field of biometrics. The ability to identify people from long distances by collecting contactless data will play an important role in solving criminal cases. IP-based camera systems are widely used in most indoor and outdoor environments today. These camera images have a resolution of 2 MP and above. Operations such as object tracking, detection and recognition by processing these images are quite costly in terms of processing load. Due to this processing load, real-time analysis of video images is difficult. In this study, relatively lower resolution data sets obtained from images containing human gait biometrics were used. Since the images in these data sets are in black/white format, it was investigated whether the lower resolution image and the original resolution image could provide similar accuracy results. Random Forests algorithm was used for this. The results obtained on the original and low-resolution images for 3 different data sets are almost the same. With this study, it has been shown that human recognition can be achieved with high accuracy in real time by reducing the resolution rate by 67.5%.
