71 results about "Haar-like features" patented technology

The embodiment of the invention provides a vehicle license plate recognition method and a vehicle license plate recognition system based on coarse positioning and fine positioning fusion. The method comprises the steps that a positive sample set and a negative sample set of a vehicle license plate image are collected, Haar-like characteristics of the positive sample set and the negative sample set are extracted to train a vehicle license plate classifier, and a vehicle license plate region in an image to be detected is subjected to coarse positioning by using the vehicle license plate classifier; the vehicle license plate region subjected to coarse positioning is converted into a gray level image, vertical and horizontal projection information is analyzed to determine the upper and lower boundaries and the left and right boundaries of a vehicle license plate, and a vehicle license plate region subjected to fine positioning is obtained; a vehicle license plate character set comprising predetermined vehicle license plate characters is collected, a tesseract engine is used for analyzing the character characteristics of the predetermined vehicle license plate characters in the vehicle license plate character set and training to obtain a vehicle license plate character repertoire, the vehicle license plate region subjected to the fine positioning is identified by the vehicle license plate character repertoire, and characters in the vehicle license plate region subjected to the fine positioning are determined. According to the vehicle license plate recognition method and the vehicle license plate recognition system based on coarse positioning and fine positioning fusion, the problem that the positioning precision of the vehicle license plate is subjected to interference due to complicated background is solved.

The invention relates to a method for identifying multi-class facial expressions at a high precision based on Haar-like features, which belongs to the technical field of computer science and graphic image process. Firstly, the high-accuracy face detection is achieved by using the Haar-like features and a series-wound face detection classifier; further, the feature selection is carried out on the high-dimension Haar-like feature by using the Ada Boost. MH algorithm; and finally, the expression classifier training is carried out by using the Random Forest algorithm to complete the expression identification. Compared with the prior art, the method can reduce training and identifying time while increasing the multi-class expression identification rate, and can implement the parallelization conveniently to increase the identification rate and meet the requirement of real-time processing and mobile computing. The method can identify the static image and the dynamic image at a high precision, is not only applicable to the desktop computer but also to the mobile computing platforms, such as cellphone, tablet personal computer and the like.

Faces in images are quickly detected with minimal memory resource usage. Instead of calculating a Haar-like feature value by subtracting the average pixel intensity value in one rectangular region from the average pixel intensity value in another, adjacent rectangular region, a face-detection system calculates that Haar-like feature value by dividing the average pixel intensity value in one such rectangular region by the average pixel intensity value in the other such adjacent rectangular region. Thus, each Haar-like value is calculated as a ratio of average pixel intensity values rather than as a difference between such average pixel intensity values. The feature values are calculated using this ratio-based technique both during the machine-learning procedure, in which the numerical ranges for features in known face-containing images are learned based on labeled training data, and during the classifier-applying procedure, in which an unlabeled image's feature values are calculated and compared to the previously machine-learned numerical ranges.

The invention discloses a face recognition based multi-camera video event retrospective trace method. The face recognition based multi-camera video event retrospective trace method comprises the steps of firstly training a face cascade classifier by adopting an Adaboost method based on a Haar-like feature, then adopting the face cascade classifier to detect and tracing faces existing in each camera so as to obtain a face library, framing face zones of pedestrians in videos, extracting the LBP features of the face zones, conducting matching recognition on target faces in the face library, finally extracting the time of targets existing in all cameras to obtain walking routes according to face recognition results so as to obtain retrospective trace results of the targets. The face recognition based multi-camera video event retrospective trace method is applied in multiple camera video system in multiple scenes and used for retrospection of video events, the working amount of manual video inquiry is reduced, and inquiry efficiency is improved.

The invention relates to a potato image acquisition device based on RGB-D cameras and a method for identifying and locating a bud eye. The image acquisition device consists of three RGB-D cameras thatacquire the color maps and the depth maps of a potato sample from three different angles. The image processing part comprises obtaining a potato target image by preprocessing the color maps and performing a mask-based target extraction method; then training a classifier by using an Adaboost algorithm and a Haar-like feature, identifying a bud eye area, and obtaining the two-dimensional coordinates of the bud eye area; calibrating the RGB-D cameras to obtain the internal parameters and the external parameters of respective RGB-D cameras, and generating point cloud in combination with the depthmaps and the color maps acquired by the corresponding cameras; successively registering the three groups of point cloud by using an iterative nearest neighbor algorithm to obtain the three-dimensional model of the potato sample; then converting the obtained two-dimensional coordinates of the bud eye into coordinates in a three-dimensional space where the three-dimensional model of the potato sample is located, so as to achieve the three-dimensional positioning of the potato bud eye and lay a foundation for realizing the automation of the potato seed dicing.

The invention relates to a target tracking method and a system based on an on-line initialization gradient enhancement regression tree. In the system consisting of a video input end, a tracking target output end and an on-line training classifier, the method comprises the steps of 1) selecting a tracking target from a video series and extracting positive and negative samples of a Haar-like feature, 2) randomly establishing the on-line classifier according to the positive and negative samples to obtain a training residual error, 3) conducting training amendment by taking the training residual error as a training sample of the on-line classifier and establishing a target model, and 4) acquiring an image confidence map from a next frame of video image, determining a maximum position of a confidence value in a target window, and accomplishing tracking. According to the method and the system, the tree can be converged to an optimum point quickly to ensure that the random forest detection optimization is accomplished, the classifier is updated through on-line study, and the problems of appearance variation, rapid movement and shielding of a target are solved well.

The invention discloses a training device, a training method and a detection method. The training device comprises a sample acquiring unit, a feature extracting unit and a training unit, wherein the sample acquiring unit is used for acquiring an object image set and a background image set which have preset numbers; the feature extracting unit is used for extracting features of the object image setand the background image set acquired by the sample acquiring unit; the training unit is used for carrying out training by using the features extracted by the feature extracting unit so as to acquirea classifier; the features comprise a wavelet-like feature and a gradient orientation histogram feature; and the classifier adopts a specific cascading structure, combines a Haar-like feature with aHOG feature together for use, plays the advantages of quick calculation of the Haar-like feature and strong resolution of the HOG feature, and simultaneously improves the performance of the acquired classifier greatly without increasing the calculating cost obviously.

An interaction robot intelligent motion detection and control method based on a nerve network uses a conventional characteristic engineering and classifier as a human eye portion for primary positioning, the nerve network can carry out behavior high efficiency identification, and a limit state machine is introduced to realize state fast conversion. Stage 1: extracting Haar-like features and combining with an Adaboost cascade classifier to finish interaction robot face identification and face human eye identification, and using a convolution nerve network to carry out human eye secondary detection and screening; stage 2: calculating a dual-eye light flow graph, using CNN to extract spatial domain characteristics, using LSTM to extract time domain characteristics, and finishing human eye behavior identification; stage 3: using the limit state machine FSM to finish state conversion. The system is high in detection precision, and fast in state conversion.

The invention discloses a workpiece detecting and sorting device and method based on a visual robot. The device comprises a transmission device, a robot and a computer. The transmission device collects a workpiece image, analysis is carried out through a computer, and the robot grabs a damaged workpiece. Compared with the prior art, the workpiece detecting and sorting device and method based on the visual robot is characterized in that Haar-like features are adopted to establish a weak classifier based on gray level changes, an AdaBoost algorithm is used for forming a strong classifier with high accuracy, cascading of the strong classifier is then carried out, a region of interest (ROI) containing a target object is extracted in the image, and the better detection accuracy and the real-time effect are obtained; a defect image is obtained in the target object, and a level set method is used for obtaining a defect ROI to form a training sample; and a convolutional neural network (CNN) isintroduced as a training model, and the accuracy of defect classification is improved.

The invention relates to an on-line type visual tracking method which comprises the following steps. Firstly, a classifier is initialized: a first frame image is obtained and a to-be-tracked object region is marked; analogous hal features of an image block of the object region can be calculated and a first classifier is initialized; binary coding features of the image block of the object region can be calculated and the second classifier can be initialized. Secondly, self-adaptive tracking: based on a support vector machine which is output structurally, tracking of a particle filter can be combined and a motion model is built. Thirdly, a target can be captured renewably: a first classifier can be used for renewing a training sample and then, the second classifier can be used for building an overall situation gird so as to implement a global search. According to on-line type visual tracking method, the two classifiers can be used: the first classifier can be used for the self-adaptive tracking; the second classifier can be used for capturing lost targets renewably. Problems that target appearance changes and the targets disappear and need obtaining renewably can be respectively solved. A mass of practices cannot need to be proceeded before tracking targets.

The invention discloses an SAR (Synthetic Aperture Radar) image ship target identification method based on multi-level feature deep fusion. The SAR image ship target identification method comprises the following steps: (1) processing an SAR image sample by adopting a Haar-like feature template and reducing dimensions to obtain low-level Haar-like features; (2) processing the SAR image sample by using a convolutional neural network to obtain high-level depth features; (3) fusing the low-level Haar-like features and the high-level depth features by using a multi-level deep learning network to obtain a multi-level feature weight coefficient, and then obtaining an optimal SVM classifier through learning training; and (4) identifying the input SAR image sample slice to be identified by using the multi-level feature weight coefficient and the SVM classifier. According to the method, the detection performance of the SAR image ship target can be effectively improved, and the method has high engineering application value.

The invention discloses an RoIs (regions of interest) filtering method and device for vehicle-mounted thermal imaging pedestrian detection. The method provided by the invention is a method for filtering non-pedestrian RoIs through a three-level cascading filter, and comprises the steps that the first layer calculates a pedestrian pixel height and the RoIs aspect ratio and sets corresponding threshold intervals to filter out the RoIs in abnormal sizes; the second layer calculates the vertical distances between the upper and lower boundaries of each RoI and a current image pavement reference oneby one, calculates the threshold value based on the threshold of the pixel heights of the RoIs, and filters out the RoIs in abnormal sizes; the third layer searches for a possible pedestrian head region according to the luminance vertical projection difference curve of each RoI, performs the comparison of the difference degree of the Haar-like features of a head region and an adjacent backgroundregion, and filters out the RoIs without the pedestrian heads. The method can reduce the calculation expenditure of the pedestrian detection and improves the scene adaptability of a classifier under the condition that the pedestrian detection accuracy is considered. The RoIs filtering device comprises an abnormal size RoIs filter, an abnormal position RoIs filter and a head missing RoIs filter.

A method identifies an unknown face in an input image using reference images of known faces. A Haar-like feature vector is extracted from each image. The vectors are compressed. An L1 norm is determined between the compressed feature vector of the input image and each compressed feature vector from the set of reference images to determine a most similar reference image. The identity of the face associated with the most similar reference image is assigned as the identity of the unknown face in the input image.

The invention discloses a vehicle detection system which can be divided into an off-line training module and an on-line identification module. For the off-line training module, through learning of a large amount of vehicle samples or non-vehicle samples acquired in various environments, feature distribution of searching-type weak classifier stimulation samples is designed according to haar-like features of samples, and a plurality of weak classifiers are automatically selected from a weak classifier space through a continuous Adaboost algorithm to be combined into a strong classifier. According to the video vehicle detection method based on continuous Adaboost, the Adaboost algorithm is improved and made to be capable of processing a classifier with continuous reliability output, and the Adaboost algorithm can converge faster.

The present invention discloses a pedestrian detection method based on Haar-like intermediate layer filtering features, comprising the steps of: extracting object features of various training images in a training image set, training an Adaboost classifier based on decision trees by using the extracted object feature data to obtain a classification model; extracting object features of an image to be detected under a plurality of scales and inputting the object features to the classification model to obtain a pedestrian detection result, wherein a method for extracting the object features comprises the following steps: respectively extracting a plurality of different channel features of an original image to obtain multiple channel feature patterns of the original image; respectively performing downsampling for each channel feature pattern; respectively extracting corresponding Haar-like features of each channel feature pattern which has been subjected to the downsampling by using a group of preset Haar-like feature templates; and clustering all the Haar-like features of the original image into the object features of the original image. The invention also discloses a pedestrian detection apparatus based on Haar-like intermediate layer filtering features. The pedestrian detection method and the pedestrian detection apparatus can effectively improve pedestrian detection performance.

The invention provides a passenger flow counting method. In view of the feature that an image photographed at the passenger station non entrance-exit area is likely to be blocked, head and shoulder haar-like features which are not likely to be blocked and whose forms remain basically unchanged of the passenger are raised to detect the passenger, the accuracy of detecting whether to be the passenger is high, and the method is applicable to an application scene in which the passenger image is blocked. After detection, through a Kalman filter and the above passenger detection method, double tracking is carried out on the position of each frame of image of the passenger, and thus, the tracking accuracy is ensured. Experiments prove that the method can achieve good tracking effects in view of features that poses of passengers at the passenger station non entrance-exit area are changeable, behaviors are complicated, and the walking directions are hard to predict. Through accurately detecting and tracking the passengers, the method provided by the invention can effectively record the movement track of the passenger, passenger target matching is ensured, error detection and missing detection are reduced, and the person number counting precision is improved.

The invention discloses a Real Sense-based facial expression animation driving method, and relates to the field of computer graphics. The method comprises the following steps of: 1) obtaining a depthimage and preprocessing the depth image to obtain a depth information greyscale map; 2) extracting samples on the basis of the depth information greyscale map to obtain a depth information training set, extracting a Haar-like feature set from the depth information training set, carrying out Adaboost training to obtain a cascade face classifier and tracking a face position; 3) mapping the tracked face position to a color image to obtain a color image face position, and extracting face features through an algorithm; and 4) processing the face features and the color image face position, and matching the face features and the color image face position with a face model AU so as to complete expression animation driving. According to the method, the problems of low precision, low efficiency andcomplicated processing as existing virtual facial expression display technology for performing driving adopts a common camera to realize capture and feature extraction and is not suitable for complicated backgrounds are solved, and the effect of improving the facial expression animation driving efficiency, correctness and robustness is achieved.

The invention provides an SVM object detection method based on Adaboost Haar-Like features. According to the method, Haar-Like features are adopted, a plurality of features of the type constitute a strong feature classifier through an Adaboost method, and multiple strong feature classifiers are selected through the method; and then, the feature values of the strong feature classifiers constitute a feature vector according to a certain sequence, and the feature vector is trained and detected by an SVM method. Although the method of the invention is of lower speed than an object detection algorithm based on cascaded Adaboost Haar-Like features, the accuracy rate is increased obviously.

The invention relates to a target tracking method based on sparse feature selection. According to the method, firstly, a target, a background and a to-be-selected target point are expressed by use of Haar-like features; then, feature selection is performed on high-dimensional Haar-like features by use of a special nature of sparse expression, and features well distinguishing the target and background are selected as expression of sample points; and finally, selected sample points are used for training a na ve Bayes classifier, and updating classification is performed online, so that the classifier can reflect the relation between the target and background in real time. A projection matrix is constructed by use of the sparse feature selection method, dimensionality reduction is performed on conventional high-dimensional Haar-like features, the calculated quantity is reduced, meanwhile, features helpful to classification are retained, the target and background can be distinguished more effectively, and rapid robustness tracking of the target is realized.

The invention discloses a weighted extreme learning machine video target tracking method based on weighted multi-example learning, solving the problem of bad tracking accuracy in the prior art. The method includes 1. initializing a Haar-like feature similar model pool and constructing a variety of feature model blocks, setting the weighted extreme learning machine network parameters; 2. extracting the training samples in the current frame and their feature blocks corresponding to the feature blocks of the different feature model blocks; 3. calculating the weighted multi-instance learning weight values; 4. constructing a plurality of networks corresponding to the different feature blocks and selecting the network with the largest similarity function value of the packet and the corresponding feature model block; 5. calculating the network global output weight values; 6. extracting the detection samples in the next frame and their corresponding feature blocks corresponding to the selected feature model blocks; 7. classifying the detection samples by means of the selected network and obtaining the target position of the next frame; and 8. repeating the above steps until the video is ended. According to the invention, the tracking accuracy is improved, and the target robustness tracking is realized.

The present invention discloses an online multi-instance learning-based soccer video player tracking method, and belongs to the computer visual recognition field. The technical scheme is characterized by at the target feature extraction aspect, combining the global features and the local features, extracting a place dominant color and a player template dominant color histogram; at the same time, carrying out the particle initialization on a particle filter motion model, transferring the states of all particles at the previous frame of target player positions, calculating the similarity of all particles and the player template dominant color histogram after the state transfer, removing the influence of the place dominant color, carrying out the normalization on the particle weights according to the similarity value, and using the particles of large weights to substitute to generate a new particle set; obtaining the Haar-like feature vectors of a set image, inputting in a multi-instance learning classifier, and calculating the current frame of target player positions. According to the technical scheme of the present invention, the nondeterminacy of the target motion can be reduced, a drift phenomenon in the tracking is inhibited effectively, and the accuracy of a tracking result is improved.