Logo multitel

Costal surveillance

Video Monitoring of coastal areas has increased, during latest years, in terms of number of deployed stations and type of technology and approaches used. At Multitel, thanks to our computer vision expertise and background, we undertake this topic mainly in order to monitor and manage coastal zones.

Main research and development investigated are related to the sea segmentation (also known as shoreline detection), object-oriented extraction such as people detection for beach occupancy measurement, or dune morphology monitoring… The idea is to be able to provide public authorities with useful and relevant about coastal zones, for environmental or operational considerations.

Sea segmentation

Sea segmentation is based on a statistical color modeling using a sparse dictionary representation. Pixels are classified regarding the distance measurement against the dictionary color representatives.

Color dictionary is incrementally learned from user inputs. Colors features are extracted from user defined area using smart clustering algorithms.

Therefore, the dictionary is iteratively enriched with new color features, however, a pruning algorithm discard less relevant features to maintain an ideal dictionary size for enhanced performances and reduced computation time.

Below can be found some examples of the obtained results in four different contexts.

Original Images

Result Images

People detection

Humans are detected in the images by a supervised learning algorithm:

  • A person appearance model has been learned automatically from a large collection of image examples.
  • The model uses a complex combination of shape and color features, and is trained to optimally discriminate between a person and any other background object.
  • The model is then scanned on the image at every scales and positions in order to find the people localizations.

Despite the complexity of the model, several optimizations allow to keep a low computation time, compatible with real time applications.

Below can be found some examples of the obtained results in four different contexts.

Parasol detection

To detect parasols on the beach, the designed algorithm uses the color of the parasols. Indeed, on private beach the parasols have always the same color (yellow and blue on the following images).

The designed algorithm is in four steps:

  1. Extract the target color using a mixture of gaussians.
  2. Cluster each pixel of the image using this mixture.
  3. Detect the different connected components.
  4. Filter them based on different criteria (like size) to obtain the detections.

On the following images can be found some examples of the obtained results in two different contexts.