Fastest object detection module?

Hi, Im working with the object detection with bird eye viewer sample code. I need the fastest processing time as possible for the while loop to get real-time measurement of human position in the world. So far my computation time varies between 0.16 to 0.2 seconds. Any way I can improve this? Im using the zed2 camera and the zedbox. Here are my configuration for the camera and for object detection module:

Set configuration parameters

init_params = sl.InitParameters()
init_params.camera_resolution = sl.RESOLUTION.HD720
init_params.camera_fps = 60
init_params.coordinate_system = sl.COORDINATE_SYSTEM.RIGHT_HANDED_Z_UP # Use a right-handed Y-up coordinate system
init_params.coordinate_units = sl.UNIT.METER # Set units in meters
init_params.depth_mode = sl.DEPTH_MODE.ULTRA
init_params.sdk_verbose = True

obj_param = sl.ObjectDetectionParameters()
obj_param.detection_model = sl.DETECTION_MODEL.HUMAN_BODY_FAST
obj_param.image_sync = True
obj_param.enable_tracking = True
obj_param.enable_body_fitting = True

obj_runtime_param = sl.ObjectDetectionRuntimeParameters()
obj_runtime_param.detection_confidence_threshold = 40

objects = sl.Objects() #preparing the zed objects variable
runtime_parameters = sl.RuntimeParameters()
runtime_parameters.measure3D_reference_frame = sl.REFERENCE_FRAME.WORLD

thak you beforehand.

You can change the sl.DEPTH_MODE from ULTRA to PERFORMANCE in order to gain more computional power. However, it can decrease the quality of the data you receive.

  • init_params.depth_mode = sl.DEPTH__MODE.ULTRA => sl.DEPTH__MODE.PERFORMANCE

Benjamin Vallon

Stereolabs Support