Method and system for recognizing the intentions of shoppers in retail aisles based on their trajectories

• US 9,740,977 B1
• Filed: 05/29/2009
• Issued: 08/22/2017
• Est. Priority Date: 05/29/2009
• Status: Active Grant

• Current Assignee: Videomining Corporation
• Original Assignee: Videomining Corporation
• Inventors: Moon, Hankyu, Sharma, Rajeev, Jung, Namsoon
• Primary Examiner(s): Crawley, Talia
• Application Number:
  US12/455,295
• Time in Patent Office: 3,007 Days
• Field of Search: 705 729, 382181, 379 8802, 701200
• CPC Class Codes: G06N 5/00 Computing arrangements using knowledge-based models
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06T 7/00 Image analysis
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69

Abstract

The present invention is a method and system for automatically recognizing which products a shopper intends to find or purchase based on the shopper'"'"'s trajectory in a retail aisle. First, the system detects and tracks the person to generate the trajectory of the shopper. Then some of the dynamic features are extracted from the shopper trajectory. The shopper trajectory features of a given trajectory are typically the positions, the motion orientations, and speeds at each point of the trajectory. A shopper behavior model is designed based on some of the primitive actions of shoppers. The last step of the method is to analyze a given shopper trajectory to estimate the shopper'"'"'s intention. The step either utilizes decision rules based on the extracted shopper trajectory features, or utilizes a trained Hidden Markov Model, to estimate the progression of the primitive actions from the trajectory. The decoded progression of the shopper behavior states is then interpreted to finally determine the shopper'"'"'s intention.

Claims

1. A method for determining a shopper'"'"'s intention in an aisle, based on the shopper'"'"'s trajectory, comprising the following steps of:

a) providing a hardware comprising at least a camera, a plurality of machine learning-based human body detectors or a multi-hypothesis tracker, a computer processor and a set of computer vision algorithms wherein the camera captures a plurality of images, and the plurality of machine learning-based human body detectors or a multi-hypothesis tracker detects a set of people across the plurality of images, and wherein the computer processor tracks the shopper trajectories of the set of people, extracts shopper trajectory features, and analyzes the shopper trajectory features for shopper behaviors, wherein the hardware determines the shopper'"'"'s intention in the aisle by the steps of;

b) setting up a shopper behavior model,

wherein the shopper behavior model consists of a relationship between shopper trajectory dynamics and atomic behaviors,

wherein the shopper trajectory dynamics comprise position, speed, and orientation,

c) detecting and tracking the shopper from input images captured by the camera to generate a trajectory of the shopper,

d) converting the shopper trajectory into a lateral trajectory wherein the lateral trajectory into y and t coordinates, wherein y coordinates are in the direction perpendicular to an aisle and t coordinates are instances of time,

e) extracting shopper trajectory features from the trajectory, wherein the shopper trajectory features include spatiotemporal dimensions, and

f) analyzing the shopper trajectory features of the shopper'"'"'s trajectory based on the shopper behavior model to determine the shopper'"'"'s behavior state,

wherein the lateral direction is a direction perpendicular to the aisle direction, and

wherein the shopper trajectory features comprise lateral position, lateral position histogram, motion orientation, and motion speed.

2. The method according to claim 1, wherein the method further comprises a step of utilizing proximity to a nearest shopper as a part of the shopper trajectory features,

wherein the shopper behavior model is augmented with a state for avoidance between shopper trajectories, wherein the state of avoidance distinguishes avoiding another shopper from another shopping behavior comprising browsing or approaching.

3. The method according to claim 1, wherein the method further comprises a step of constructing the lateral position histogram by dividing the lateral distance between shelves in an aisle into bins of an equal size,

wherein the number of lateral trajectory instances is counted for each bin, wherein the number of lateral trajectory instances in each bin indicates the amount of time spent at the lateral position, and wherein the number of lateral trajectory instances in each bin identifies the shopper behavior.

4. The method according to claim 1, wherein the method further comprises a step of setting up the shopper behavior model with states for walking, browsing, approaching, interaction, and avoidance as behavioral primitives of the shopper behavior model,

wherein each state of the shopper behavior model is associated with typically observed shopper trajectory features, and

wherein the shopper trajectory features include a lateral position of a shopper.

5. The method according to claim 1, wherein the method further comprises a step of setting up the shopper behavior model with states for left-browsing, right-browsing, left-approaching, right-approaching, left-interaction, and right-interaction as primitives of the shopper behavior model, wherein the primitives distinguish between two similar trajectories with different shopper intentions,

wherein each state of the shopper behavior model is associated with typically observed shopper trajectory features, and

wherein the observed shopper trajectory features are differentiated between left states and right states.

6. The method according to claim 1, wherein the method further comprises the following steps of:

a) annotating training shopper trajectories with ground truth shopper behavior states,

b) extracting training shopper trajectory features from the training shopper trajectories, and

c) training a Hidden Markov Model (HMM) framework using the Baum-Welch HMM encoding algorithm with the training shopper trajectory features.

7. The method according to claim 1, wherein the method further comprises a step of using a rule-based model to analyze the shopper trajectory features,

wherein decision rules in the rule-based model are constructed based on an initial lateral position between shelves in the aisle, relative positions of quartile points in the lateral position histogram, and a destination information derived from a dominant motion orientation and an average motion speed.

8. The method according to claim 1, wherein the method further comprises a step of computing a level of entry from a sequence of lateral positions, quartile points from the lateral position histogram, a dominant motion orientation, and an average motion speed.

9. An apparatus for determining a shopper'"'"'s intention in an aisle, based on the shopper'"'"'s trajectory, comprising:

a) providing a hardware comprising at least a camera, a plurality of machine learning-based human body detectors or a multi-hypothesis tracker, a computer processor and a set of computer vision algorithms wherein the camera captures a plurality of images, and the plurality of machine learning-based human body detectors or a multi-hypothesis tracker detects a set of people across the plurality of images, and wherein the computer processor tracks the shopper trajectories of the set of people, extracts shopper trajectory features, and analyzes the shopper trajectory features for shopper behaviors, wherein the hardware determines the shopper'"'"'s intention in the aisle by the steps of;

b) setting up a shopper behavior model,

wherein the shopper behavior model consists of a relationship between shopper trajectory dynamics and atomic behaviors,

wherein the shopper trajectory dynamics comprise position, speed, and orientation,

c) detecting and tracking the shopper from input images captured by the camera to generate a trajectory of the shopper,

d) converting the shopper trajectory into a lateral trajectory wherein the lateral trajectory into y and t coordinates, wherein y coordinates are in the direction perpendicular to an aisle and t coordinates are instances of time,

e) extracting shopper trajectory features from the trajectory, wherein the shopper trajectory features include spatiotemporal dimensions, and

f) analyzing the shopper trajectory features of the shopper'"'"'s trajectory based on the shopper behavior model to determine the shopper'"'"'s behavior state,

wherein the lateral direction is a direction perpendicular to the aisle direction, and

wherein the shopper trajectory features comprise lateral position, lateral position histogram, motion orientation, and motion speed.

10. The apparatus according to claim 9, wherein the apparatus further comprises a computer for utilizing proximity to a nearest shopper as a part of the shopper trajectory features,

wherein the shopper behavior model is augmented with a state for avoidance between shopper trajectories, wherein the state of avoidance distinguishes avoiding another shopper from another shopping behavior comprising browsing or approaching.

11. The apparatus according to claim 9, wherein the apparatus further comprises a computer for constructing the lateral position histogram by dividing the lateral distance between shelves in an aisle into bins of an equal size, wherein the number of lateral trajectory instances is counted for each bin, wherein the number of lateral trajectory instances in each bin indicates the amount of time spent at the lateral position, and wherein the number of lateral trajectory instances in each bin identifies the shopper behavior.

12. The apparatus according to claim 9, wherein the apparatus further comprises a computer for setting up the shopper behavior model with states for walking, browsing, approaching, interaction, and avoidance as behavioral primitives of the shopper behavior model,

wherein each state of the shopper behavior model is associated with typically observed shopper trajectory features, and

wherein the shopper trajectory features include a lateral position of a shopper.

13. The apparatus according to claim 9, wherein the apparatus further comprises a computer for setting up the shopper behavior model with states for left browsing, right-browsing, left-approaching, right-approaching, left-interaction, and right-interaction as primitives of the shopper behavior model, wherein the primitives distinguish between two similar trajectories with different shopper intentions,

wherein each state of the shopper behavior model is associated with typically observed shopper trajectory features, and

wherein the observed shopper trajectory features are differentiated between left states and right states.

14. The apparatus according to claim 9, wherein the apparatus further comprising a computer that performs the following steps of:

a) annotating training shopper trajectories with ground truth shopper behavior states,

b) extracting training shopper trajectory features from the training shopper trajectories, and

c) training a Hidden Markov Model (HMM) framework using the Baum-Welch HMM encoding algorithm with the training shopper trajectory features.

15. The apparatus according to claim 9, wherein the apparatus further comprises a computer for using a rule-based model to analyze the shopper trajectory features,

wherein decision rules in the rule-based model are constructed based on an initial lateral position between shelves in the aisle, relative positions of quartile points in the lateral position histogram, and a destination information derived from a dominant motion orientation and an average motion speed.

16. The apparatus according to claim 9, wherein the apparatus further comprises a computer for computing a level of entry from a sequence of lateral positions, quartile points from the lateral position histogram, a dominant motion orientation, and an average motion speed.