2 results about "Temporal discretization" patented technology

This application relates to a spatiotemporal discretization control circuit and control method. The spatiotemporal discretization control method includes a spatial discretization mode, comprising: acquiring a spatial configuration value, which represents the workload required for the control circuit of the power consumption unit array to output; acquiring a spatial grouping value; which is used to divide the workload into 2... N The application involves grouping the workload into sub-workloads; obtaining the channel address value; dividing the channel address value into two parts based on the spatial grouping value; dividing the spatial configuration value into two parts based on the spatial grouping value; and comparing the second part of the bit-reversed channel address value with the first part of the spatial configuration value. This application also relates to a spatiotemporal discretization control method, including a time discretization mode. This application further relates to a spatiotemporal discretization control circuit, including a spatial discretization module. This application also relates to a spatiotemporal discretization control circuit, including a time discretization module.

The application discloses a kind of multi-stage space-time modeling road target detection methods based on event data.The present application constructs a kind of target detection network of fusion spatial feature modeling, space modulation context broadcast (SM-CB) module, deep separable convolution gated unit (DSConvGRU) module, pulse timing memory (STM) module and feature fusion and detection module for the problems of strong event camera data sparsity, time sequence dependence is significant and the lack of space-time information modeling of existing method.The method is first time discretization and space accumulation to event stream data, generates structured event representation;Subsequently in multi-stage feature extraction process, realize spatial feature modeling by local and global self-attention mechanism;On this basis, introduce SM-CB module to aggregate and space modulation global semantic information, enhance feature expression ability;While combining DSConvGRU module and STM module, the dynamic change of event data is modeled;Finally, through multi-scale feature fusion and detection head, the detection and positioning of target are realized.The present application can effectively improve the precision and robustness of pedestrian and vehicle target detection in complex road scene.