Abstract:Aiming at the problem of insufficient robustness ofmoving target tracking in complex scenes

a particle filter target trackingalgorithm based on CNN feature extraction is proposed CNN selflearning mechanism is used toextract highlevelsemantic features of objects in images Chaotic sequence variable scale fireflyalgorithm is introduced to improve the accuracy of target recognition Aparticle filter tracking algorithm combining mean shift and weight optimizationis constructed to optimize the weight of particles

improve the diversity ofparticle sets and make video tracking more accurate The simulation results showthat the proposed algorithm can effectively adapt to occlusion

illumination

violent motion and other scenes

and has good adaptability and high realtime performance to illuminationchange

scale change

occlusion and so on The results of comparison with sevenother methods show that under the same experimental conditions

the trackingsuccess rate and tracking accuracy of this method are 5%~40% higher than thoseof other methods