Aiming at the problem that the target can not be re-locked after the target face is lost due to the interference, an optimized tracking algorithm is proposed to track the human face in the video for a long time. Based on KCF (Kernel Correlation Filter) and Camshift (Continuously Adaptive Mean-Shift) algorithm, the "voting" module is ad the foreground area, Camshift tracking algorithm and KCF tracking algorithm, the algorithm of tracking failure is determineded. By comparingd. Re-lock automatically. The experimental results show that the proposed algorithm can achieve stable tracking of the target, and automatically re-lock and continue to track the target with the track, to achieve the purpose of tracking the target for a long time.

Using text question and answer technology to improve the efficiency of judicial personnel in the process of case handling can greatly reduce labour costs. This paper proposes a hybrid neural network model that combines pre-training model BERT and Bi-GRU. The model first uses the pre-training model to learn powerful semantic capabilities, then combines Bi-GRU to learn the semantic information between the text and the question, finally gets the answer corresponding to the legal text. The experimental results on the CJRC data set show that compared with the basic baseline model, the algorithm in this paper can effectively improve the accuracy and F1 value.