41 results about "Background factors" patented technology

The invention discloses a pedestrian re-identification method, including pedestrian feature extraction and feature measurement; the pedestrian feature extraction mainly uses a sliding window to extract the image color histogram, expands the main color, and counts each color in each row of sliding windows The number of times the pattern appears, select the larger sum as the feature output of the color, the sliding window traverses the entire image, and forms the feature after normalization. In order to make the matching more universal, the features of the search source from the whole body to the half body are extracted. When calculating the measurement, the Euclidean distance between the feature dimension and the target feature to be searched is made with the feature dimension of the search image feature vector not being 0, which further reduces the interference of background factors in complex scenes; the pedestrian re-identification method provided by the present invention is applied to The re-identification of pedestrians in complex scenes is highly portable to scene changes, the algorithm is stable and fast, and it can effectively solve the problem of low quality of surveillance video and databases to be searched, and has strong practicability.

The invention discloses an image processing method and an image processing device. The image processing method is used for processing mixture of source image pixel points and background pixel points, and comprises the steps of: acquiring a source image alpha value, a source image global alpha value and a source image color value of the source image pixel points; selecting the source image alpha value, the source image global alpha value, or a product value of the source image alpha value and the source image global alpha value as a source image calculation value according to a source image blending mode value; selecting 0, 256, the source image calculation value, a difference value between 256 and the source image calculation value, or a background estimated value as a background factor value according to a source image blending factor value; and calculating a source picture color calculation value according to the source image color value and the source image calculation value. The image processing method integrates mix_alpha and port_duff alpha calculation algorithms, simplifies logic corresponding to the two different algorithms, reduces area required by alpha operating hardware, and improves power utilization rate.

The invention relates to a pedestrian clothing attribute recognition method based on depth pose estimation and multi-feature fusion. The method first selects some retrieval results for subsequent attribute recognition through appearance feature matching; then through the SSD-based deep human pose estimation method, it can effectively locate the foreground area belonging to pedestrians in the image, and better eliminate the interference of background factors Finally, the analysis results of various methods are combined, combined with the iterative smoothing process, the maximum a posteriori probability distribution method is adopted to strengthen the correlation between attribute labels and pixels, and the final attribute analysis and recognition results are obtained. The invention solves the problems of inaccurate label identification and pixel analysis area deviation under a single analysis mode. The method is simple and flexible, and has strong practical applicability.

This invention discloses a multi-channel active controller, which comprises an error signal conditioner, a digital signal processor, a filter, a control signal driver and reference signal conditioner. Wherein, the signal from error sensor located outside the multi-channel active controller sequentially passes error signal conditioner, digital signal processor, filter and control signal driver; outer reference signal via reference signal condition is input to digital signal processor consisting of a plurality of A / D converters, real-time processor chips, and background factor updating data processor chips. The invention is operated simply, can maintain reality of time and use the controller traceability.

The invention discloses a background modeling method based on sample local density outlier detection. Calculate the local background factor of the observed new pixel value; compare the local background factor of the newly observed new pixel value with its nearest neighbor sample points to determine whether it belongs to the background, and if it belongs to the background, integrate the new pixel value into The background model is updated, and the sample point with the largest local background factor in the sample set is replaced by the new pixel value; the present invention uses the observed pixel real value to initialize the background model, which increases the authenticity of the background model and avoids being incorrectly The presence of false values ​​affects the possibility.