Face Detection Algorithm Based On Neural Network Under Complex Conditions

A key aspect of computer vision research is face detection.Facial features are widely used in detection systems,access control and security.Face detection and recognition is one of the most popular research directions in the security field.Face detection tasks become arduous due to exposure,low resolution,extreme deflection of face angles,facial occlusion,etc.,and these problems are also widespread in practical applications.In the early days,Viola and Jones invented the VJ face detection algorithm,which has been able to detect positive face images very well,but for various expressions of face,the intensity of light,various angles of deflection,etc.,The VJ detector is not ideal in practical applications.Due to the continuous research and development of convolutional neural networks,and the acceleration of computing power by hardware such as GPUs,the development of computer vision has become more and more rapid.Convolutional neural networks have incomparable advantages in the research direction of face detection,such as greatly reducing the labor cost,the characteristics of self-learning faces,and the adaptability to faces under complex conditions.However,face detection and alignment in lighting and occlusion,various postures,and unconstrained environments are challenging.Recent research has found that deep learning algorithm can solve the above problems well.Based on the theoretical knowledge of deep learning,this paper makes a deep discussion on the problem of face detection.The advantages and disadvantages of some early and existing face detection algorithms are analyzed and summarized,And then we analyzes and improves the existing convolutional neural network model,and proposes a deep-cascading multi-tasking framework to improve the performance by using the inherent relationship between detection and calibration.In particular,we use a three-tiered cascading framework combined with an improved algorithm for convolutional neural networks to detect faces and locate key points.The complexity of the improved cascaded neural network increases step by step,and each level adds a layer of convolution and pooling layers to the upper level.Since the depth of each network is quite small,the time taken for each network detection is very short,and finally the detection speed of the cascaded network is relatively fast.And the use of multi-tasking links makes the detection rate of the model higher and more accurate.Experiments show that on the deep learning framework,Tensorflow,this method is very good in the FDDB and WIDER FACE libraries which is challenge for face detection,and the average accuracy is very high,the detection time is very short.