Live demonstration: Motion detection vision sensor with dynamic background rejection

Abstract

In surveillance and monitoring applications, motion detection using commercial cameras with off-line high-level algorithm processing is not efficient. They have to force the high-level processor continuously analyze the acquired high-resolution images. This causes a large waste of power, since there are no events in most of the cases. Embedding low-level image processing on-chip will make the system to be more energy efficient. In this demonstration, we present a QVGA vision sensor embedding a low-power dynamic background subtraction algorithm [1]. The sensor detects anomalous motion and generates an alert event map as input for high-level processor. Several sensors reported in literatures detect motion events based on frame difference technique embedded on-chip, however they cannot suppress noisy zones of the scene such as swaying vegetation. Different from our previous fully analog implementation [2]-[3], this digital approach allows motion detection over a large range even in harsh outdoor scenarios. The chip consumes 1.6mW when operating at 15fps dispatching QVGA gray-scale image and event map. In this demonstration, we use FPGA module to control the vision sensor, for low power surveillance and monitoring application, it could be simply replaced with one low power processor to build a low power system which lasts for months with battery.