Abstract: A video condensation & recognition method and a system thereof are used to detect objects in input frames, derive a 3D object trajectory from which one or multiple 2D trajectories with object tubes are extracted, and catch a time position that refers to non-overlapping condition of object tubes for rearrangement of several continuous object tubes and creation of a condensed video. The generated condensed video can be adjusted by the user in the range of longest detected object tube to the generated condensed video length that is very beneficial for video analysis process. Moreover, a searched attribute of an object in a condensed video is classified for the filtering process which is able to display an object correlated with a distinct searched attribute on the condensed video.

Abstract: A video condensation & recognition method and a system thereof are used to detect objects in input frames, derive a 3D object trajectory from which one or multiple 2D trajectories with object tubes are extracted, and catch a time position that refers to non-overlapping condition of object tubes for rearrangement of several continuous object tubes and creation of a condensed video. The generated condensed video can be adjusted by the user in the range of longest detected object tube to the generated condensed video length that is very beneficial for video analysis process. Moreover, a searched attribute of an object in a condensed video is classified for the filtering process which is able to display an object correlated with a distinct searched attribute on the condensed video.

Abstract: A feature determination apparatus and method adapted to multiple object sizes are provided. The apparatus individually supplies each of the object images to a convolution neural network having several convolution layers to generate multiple feature maps corresponding to each object image. The apparatus calculates a feature amount of each feature image of each object image. The apparatus determines an invalid layer start number of each object image according to a preset threshold and the feature amount corresponding to each object image. The apparatus determines a feature map extraction recommendation for each of a plurality of object sizes according to a size of each object image and the invalid layer start number of each object image.

Abstract: A feature determination apparatus and method adapted to multiple object sizes are provided. The apparatus individually supplies each of the object images to a convolution neural network having several convolution layers to generate multiple feature maps corresponding to each object image. The apparatus calculates a feature amount of each feature image of each object image. The apparatus determines an invalid layer start number of each object image according to a preset threshold and the feature amount corresponding to each object image. The apparatus determines a feature map extraction recommendation for each of a plurality of object sizes according to a size of each object image and the invalid layer start number of each object image.

Abstract: An apparatus for monitoring an object in a low light environment includes an image capturing unit, a vehicle speed unit, an image recognizing unit, and an object determining unit. The image capturing unit continuously captures and outputs time-sliced images. The vehicle speed unit detects and outputs current vehicle speed information. The image recognizing unit recognizes a region having pixel brightness higher than a threshold in each of the time-sliced images and marks the region as a high brightness block. The object determining unit selects at least two successive time-sliced images having high brightness blocks in a continuous corresponding variation relationship from the time-sliced images, generates and outputs estimated speed information, and when the estimated speed information is different from the current vehicle speed information, determines that the high brightness block is a moving object block and monitors the moving object block.

Abstract: A wall lamp includes a positioning disk, a plurality of fastening members, a lamp shade, a connecting disk, and a retaining disk. Thus, the connecting disk is locked onto and detached from the positioning disk by rotating the connecting disk relative to the fastening members, so that the wall lamp is assembled and dismantled easily and rapidly, thereby facilitating a user assembling and disassembling the wall lamp, and thereby saving the working time. In addition, the fastening members are wholly hidden in the retaining disk, thereby enhancing the aesthetic quality of the wall lamp.