Visual Object Classification and Detection at the Edge

Visual object classification and detection can be performed at the edge with recent advances in CPU and GPU technologies.

img02
Fig.01: Multi-Sensor EO/IR payload.

The choice of AI compute can be determined by the use-case, size, weight, power and performance constraints of the platform operating at the edge.

Table.01: SWaPC comparision table
Parameter ARM v8.2 + Embedded GPU Intel i7 + Discrete GPU
Size Small form-factor (105x105x65mm) 3U chassis
Weight 0.8 to 1.2 kgs, varies with ruggedization levels 2.2 to 2.8 kgs, varies with ruggedization levels
Power 30 to 40 watts 150 to 200 watts
Performance High compute performance, depends on embedded GPU architecture Higher compute performance, depends on GPU architecture
Cost Relatively lower-cost Relatively higher-cost