Target Detection and Multi-Feature Fusion Based on Deep Learning and CamShift Algorithm

Abstract

Aiming at the problem that the continuous adaptive mean shift algorithm (Camshift) deep learning tracking algorithm based on the colour probability distribution can easily cause the tracking target to fail when the same colour interference appears in the background, we propose an improved Camshift tracking algorithm. The article first conducts target detection before Camshift tracks the target, and compares three detection algorithms through frame difference method, optical flow method, and background difference method. Then, when tracking the target, the window centre of the CamShift algorithm is compared with the centre of the target area calculated by the difference method to determine the subsequent frame search window to avoid the loss of target tracking; finally, the experiment proves that the method can effectively track the target, and it still has a very high accuracy rate when the target colour and the background colour are small.